tokencostauto 0.1.421__py3-none-any.whl → 0.1.447__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (7) hide show
  1. tokencostauto/model_prices.json +2395 -266
  2. {tokencostauto-0.1.421.dist-info → tokencostauto-0.1.447.dist-info}/METADATA +1 -1
  3. tokencostauto-0.1.447.dist-info/RECORD +9 -0
  4. tokencostauto-0.1.421.dist-info/RECORD +0 -9
  5. {tokencostauto-0.1.421.dist-info → tokencostauto-0.1.447.dist-info}/WHEEL +0 -0
  6. {tokencostauto-0.1.421.dist-info → tokencostauto-0.1.447.dist-info}/licenses/LICENSE +0 -0
  7. {tokencostauto-0.1.421.dist-info → tokencostauto-0.1.447.dist-info}/top_level.txt +0 -0
tokencostauto/model_prices.json CHANGED
@@ -407,7 +407,7 @@
407
407
  "litellm_provider": "openai",
408
408
  "max_input_tokens": 16385,
409
409
  "max_output_tokens": 4096,
410
- "max_tokens": 4097,
410
+ "max_tokens": 4096,
411
411
  "mode": "chat",
412
412
  "output_cost_per_token": 1.5e-06,
413
413
  "supports_function_calling": true,
@@ -420,7 +420,7 @@
420
420
  "litellm_provider": "openai",
421
421
  "max_input_tokens": 4097,
422
422
  "max_output_tokens": 4096,
423
- "max_tokens": 4097,
423
+ "max_tokens": 4096,
424
424
  "mode": "chat",
425
425
  "output_cost_per_token": 2e-06,
426
426
  "supports_prompt_caching": true,
@@ -432,7 +432,7 @@
432
432
  "litellm_provider": "openai",
433
433
  "max_input_tokens": 4097,
434
434
  "max_output_tokens": 4096,
435
- "max_tokens": 4097,
435
+ "max_tokens": 4096,
436
436
  "mode": "chat",
437
437
  "output_cost_per_token": 2e-06,
438
438
  "supports_function_calling": true,
@@ -446,7 +446,7 @@
446
446
  "litellm_provider": "openai",
447
447
  "max_input_tokens": 16385,
448
448
  "max_output_tokens": 4096,
449
- "max_tokens": 16385,
449
+ "max_tokens": 4096,
450
450
  "mode": "chat",
451
451
  "output_cost_per_token": 2e-06,
452
452
  "supports_function_calling": true,
@@ -460,7 +460,7 @@
460
460
  "litellm_provider": "openai",
461
461
  "max_input_tokens": 16385,
462
462
  "max_output_tokens": 4096,
463
- "max_tokens": 16385,
463
+ "max_tokens": 4096,
464
464
  "mode": "chat",
465
465
  "output_cost_per_token": 1.5e-06,
466
466
  "supports_function_calling": true,
@@ -474,7 +474,7 @@
474
474
  "litellm_provider": "openai",
475
475
  "max_input_tokens": 16385,
476
476
  "max_output_tokens": 4096,
477
- "max_tokens": 16385,
477
+ "max_tokens": 4096,
478
478
  "mode": "chat",
479
479
  "output_cost_per_token": 4e-06,
480
480
  "supports_prompt_caching": true,
@@ -486,7 +486,7 @@
486
486
  "litellm_provider": "openai",
487
487
  "max_input_tokens": 16385,
488
488
  "max_output_tokens": 4096,
489
- "max_tokens": 16385,
489
+ "max_tokens": 4096,
490
490
  "mode": "chat",
491
491
  "output_cost_per_token": 4e-06,
492
492
  "supports_prompt_caching": true,
@@ -597,7 +597,7 @@
597
597
  "litellm_provider": "text-completion-openai",
598
598
  "max_input_tokens": 16384,
599
599
  "max_output_tokens": 4096,
600
- "max_tokens": 16384,
600
+ "max_tokens": 4096,
601
601
  "mode": "completion",
602
602
  "output_cost_per_token": 1.2e-05,
603
603
  "output_cost_per_token_batches": 1e-06
@@ -608,7 +608,7 @@
608
608
  "litellm_provider": "text-completion-openai",
609
609
  "max_input_tokens": 16384,
610
610
  "max_output_tokens": 4096,
611
- "max_tokens": 16384,
611
+ "max_tokens": 4096,
612
612
  "mode": "completion",
613
613
  "output_cost_per_token": 1.6e-06,
614
614
  "output_cost_per_token_batches": 2e-07
@@ -659,7 +659,7 @@
659
659
  "litellm_provider": "openai",
660
660
  "max_input_tokens": 32768,
661
661
  "max_output_tokens": 0,
662
- "max_tokens": 32768,
662
+ "max_tokens": 0,
663
663
  "mode": "moderation",
664
664
  "output_cost_per_token": 0.0
665
665
  },
@@ -668,7 +668,7 @@
668
668
  "litellm_provider": "openai",
669
669
  "max_input_tokens": 32768,
670
670
  "max_output_tokens": 0,
671
- "max_tokens": 32768,
671
+ "max_tokens": 0,
672
672
  "mode": "moderation",
673
673
  "output_cost_per_token": 0.0
674
674
  },
@@ -677,7 +677,7 @@
677
677
  "litellm_provider": "openai",
678
678
  "max_input_tokens": 32768,
679
679
  "max_output_tokens": 0,
680
- "max_tokens": 32768,
680
+ "max_tokens": 0,
681
681
  "mode": "moderation",
682
682
  "output_cost_per_token": 0.0
683
683
  },
@@ -1063,7 +1063,7 @@
1063
1063
  "litellm_provider": "azure",
1064
1064
  "max_input_tokens": 4097,
1065
1065
  "max_output_tokens": 4096,
1066
- "max_tokens": 4097,
1066
+ "max_tokens": 4096,
1067
1067
  "mode": "chat",
1068
1068
  "output_cost_per_token": 2e-06,
1069
1069
  "supports_function_calling": true,
@@ -1076,7 +1076,7 @@
1076
1076
  "litellm_provider": "azure",
1077
1077
  "max_input_tokens": 4097,
1078
1078
  "max_output_tokens": 4096,
1079
- "max_tokens": 4097,
1079
+ "max_tokens": 4096,
1080
1080
  "mode": "chat",
1081
1081
  "output_cost_per_token": 2e-06,
1082
1082
  "supports_function_calling": true,
@@ -1368,7 +1368,7 @@
1368
1368
  "litellm_provider": "text-completion-openai",
1369
1369
  "max_input_tokens": 16384,
1370
1370
  "max_output_tokens": 4096,
1371
- "max_tokens": 16384,
1371
+ "max_tokens": 4096,
1372
1372
  "mode": "completion",
1373
1373
  "output_cost_per_token": 4e-07
1374
1374
  },
@@ -1377,7 +1377,7 @@
1377
1377
  "litellm_provider": "text-completion-openai",
1378
1378
  "max_input_tokens": 16384,
1379
1379
  "max_output_tokens": 4096,
1380
- "max_tokens": 16384,
1380
+ "max_tokens": 4096,
1381
1381
  "mode": "completion",
1382
1382
  "output_cost_per_token": 2e-06
1383
1383
  },
@@ -1661,7 +1661,7 @@
1661
1661
  "litellm_provider": "deepseek",
1662
1662
  "max_input_tokens": 131072,
1663
1663
  "max_output_tokens": 8192,
1664
- "max_tokens": 131072,
1664
+ "max_tokens": 8192,
1665
1665
  "mode": "chat",
1666
1666
  "output_cost_per_token": 1.7e-06,
1667
1667
  "source": "https://api-docs.deepseek.com/quick_start/pricing",
@@ -1818,14 +1818,13 @@
1818
1818
  "supports_tool_choice": true
1819
1819
  },
1820
1820
  "groq/gemma-7b-it": {
1821
- "deprecation_date": "2024-12-18",
1822
- "input_cost_per_token": 7e-08,
1821
+ "input_cost_per_token": 5e-08,
1823
1822
  "litellm_provider": "groq",
1824
1823
  "max_input_tokens": 8192,
1825
1824
  "max_output_tokens": 8192,
1826
1825
  "max_tokens": 8192,
1827
1826
  "mode": "chat",
1828
- "output_cost_per_token": 7e-08,
1827
+ "output_cost_per_token": 8e-08,
1829
1828
  "supports_function_calling": true,
1830
1829
  "supports_response_schema": false,
1831
1830
  "supports_tool_choice": true
@@ -2497,7 +2496,7 @@
2497
2496
  "litellm_provider": "vertex_ai-language-models",
2498
2497
  "max_input_tokens": 8192,
2499
2498
  "max_output_tokens": 2048,
2500
- "max_tokens": 8192,
2499
+ "max_tokens": 2048,
2501
2500
  "mode": "chat",
2502
2501
  "output_cost_per_character": 3.75e-07,
2503
2502
  "output_cost_per_token": 1.5e-06,
@@ -2514,7 +2513,7 @@
2514
2513
  "litellm_provider": "vertex_ai-language-models",
2515
2514
  "max_input_tokens": 8192,
2516
2515
  "max_output_tokens": 2048,
2517
- "max_tokens": 8192,
2516
+ "max_tokens": 2048,
2518
2517
  "mode": "chat",
2519
2518
  "output_cost_per_character": 3.75e-07,
2520
2519
  "output_cost_per_token": 1.5e-06,
@@ -2542,6 +2541,7 @@
2542
2541
  "supports_tool_choice": true
2543
2542
  },
2544
2543
  "gemini-1.5-pro": {
2544
+ "deprecation_date": "2025-09-29",
2545
2545
  "input_cost_per_audio_per_second": 3.125e-05,
2546
2546
  "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05,
2547
2547
  "input_cost_per_character": 3.125e-07,
@@ -2629,6 +2629,7 @@
2629
2629
  "supports_vision": true
2630
2630
  },
2631
2631
  "gemini-1.5-pro-preview-0514": {
2632
+ "deprecation_date": "2025-09-29",
2632
2633
  "input_cost_per_audio_per_second": 3.125e-05,
2633
2634
  "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05,
2634
2635
  "input_cost_per_character": 3.125e-07,
@@ -2656,6 +2657,7 @@
2656
2657
  "supports_tool_choice": true
2657
2658
  },
2658
2659
  "gemini-1.5-pro-preview-0215": {
2660
+ "deprecation_date": "2025-09-29",
2659
2661
  "input_cost_per_audio_per_second": 3.125e-05,
2660
2662
  "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05,
2661
2663
  "input_cost_per_character": 3.125e-07,
@@ -2683,6 +2685,7 @@
2683
2685
  "supports_tool_choice": true
2684
2686
  },
2685
2687
  "gemini-1.5-pro-preview-0409": {
2688
+ "deprecation_date": "2025-09-29",
2686
2689
  "input_cost_per_audio_per_second": 3.125e-05,
2687
2690
  "input_cost_per_audio_per_second_above_128k_tokens": 6.25e-05,
2688
2691
  "input_cost_per_character": 3.125e-07,
@@ -2709,6 +2712,7 @@
2709
2712
  "supports_tool_choice": true
2710
2713
  },
2711
2714
  "gemini-1.5-flash": {
2715
+ "deprecation_date": "2025-09-29",
2712
2716
  "input_cost_per_audio_per_second": 2e-06,
2713
2717
  "input_cost_per_audio_per_second_above_128k_tokens": 4e-06,
2714
2718
  "input_cost_per_character": 1.875e-08,
@@ -2743,6 +2747,7 @@
2743
2747
  "supports_vision": true
2744
2748
  },
2745
2749
  "gemini-1.5-flash-exp-0827": {
2750
+ "deprecation_date": "2025-09-29",
2746
2751
  "input_cost_per_audio_per_second": 2e-06,
2747
2752
  "input_cost_per_audio_per_second_above_128k_tokens": 4e-06,
2748
2753
  "input_cost_per_character": 1.875e-08,
@@ -2847,6 +2852,7 @@
2847
2852
  "supports_vision": true
2848
2853
  },
2849
2854
  "gemini-1.5-flash-preview-0514": {
2855
+ "deprecation_date": "2025-09-29",
2850
2856
  "input_cost_per_audio_per_second": 2e-06,
2851
2857
  "input_cost_per_audio_per_second_above_128k_tokens": 4e-06,
2852
2858
  "input_cost_per_character": 1.875e-08,
@@ -3092,7 +3098,7 @@
3092
3098
  "litellm_provider": "vertex_ai-llama_models",
3093
3099
  "max_input_tokens": 128000,
3094
3100
  "max_output_tokens": 2048,
3095
- "max_tokens": 128000,
3101
+ "max_tokens": 2048,
3096
3102
  "metadata": {
3097
3103
  "notes": "VertexAI states that The Llama 3.2 API service is at no cost during public preview, and will be priced as per dollar-per-1M-tokens at GA."
3098
3104
  },
@@ -3238,6 +3244,7 @@
3238
3244
  "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing"
3239
3245
  },
3240
3246
  "text-embedding-004": {
3247
+ "deprecation_date": "2026-01-14",
3241
3248
  "input_cost_per_character": 2.5e-08,
3242
3249
  "input_cost_per_token": 1e-07,
3243
3250
  "litellm_provider": "vertex_ai-embedding-models",
@@ -3454,6 +3461,7 @@
3454
3461
  "tpm": 4000000
3455
3462
  },
3456
3463
  "gemini/gemini-1.5-flash": {
3464
+ "deprecation_date": "2025-09-29",
3457
3465
  "input_cost_per_token": 7.5e-08,
3458
3466
  "input_cost_per_token_above_128k_tokens": 1.5e-07,
3459
3467
  "litellm_provider": "gemini",
@@ -3479,6 +3487,7 @@
3479
3487
  "tpm": 4000000
3480
3488
  },
3481
3489
  "gemini/gemini-1.5-flash-latest": {
3490
+ "deprecation_date": "2025-09-29",
3482
3491
  "input_cost_per_token": 7.5e-08,
3483
3492
  "input_cost_per_token_above_128k_tokens": 1.5e-07,
3484
3493
  "litellm_provider": "gemini",
@@ -3505,6 +3514,7 @@
3505
3514
  "tpm": 4000000
3506
3515
  },
3507
3516
  "gemini/gemini-1.5-flash-8b-exp-0924": {
3517
+ "deprecation_date": "2025-09-29",
3508
3518
  "input_cost_per_token": 0,
3509
3519
  "input_cost_per_token_above_128k_tokens": 0,
3510
3520
  "litellm_provider": "gemini",
@@ -3531,6 +3541,7 @@
3531
3541
  "tpm": 4000000
3532
3542
  },
3533
3543
  "gemini/gemini-1.5-flash-exp-0827": {
3544
+ "deprecation_date": "2025-09-29",
3534
3545
  "input_cost_per_token": 0,
3535
3546
  "input_cost_per_token_above_128k_tokens": 0,
3536
3547
  "litellm_provider": "gemini",
@@ -3556,6 +3567,7 @@
3556
3567
  "tpm": 4000000
3557
3568
  },
3558
3569
  "gemini/gemini-1.5-flash-8b-exp-0827": {
3570
+ "deprecation_date": "2025-09-29",
3559
3571
  "input_cost_per_token": 0,
3560
3572
  "input_cost_per_token_above_128k_tokens": 0,
3561
3573
  "litellm_provider": "gemini",
@@ -3598,6 +3610,7 @@
3598
3610
  "tpm": 120000
3599
3611
  },
3600
3612
  "gemini/gemini-1.5-pro": {
3613
+ "deprecation_date": "2025-09-29",
3601
3614
  "input_cost_per_token": 3.5e-06,
3602
3615
  "input_cost_per_token_above_128k_tokens": 7e-06,
3603
3616
  "litellm_provider": "gemini",
@@ -3659,6 +3672,7 @@
3659
3672
  "tpm": 4000000
3660
3673
  },
3661
3674
  "gemini/gemini-1.5-pro-exp-0801": {
3675
+ "deprecation_date": "2025-09-29",
3662
3676
  "input_cost_per_token": 3.5e-06,
3663
3677
  "input_cost_per_token_above_128k_tokens": 7e-06,
3664
3678
  "litellm_provider": "gemini",
@@ -3678,6 +3692,7 @@
3678
3692
  "tpm": 4000000
3679
3693
  },
3680
3694
  "gemini/gemini-1.5-pro-exp-0827": {
3695
+ "deprecation_date": "2025-09-29",
3681
3696
  "input_cost_per_token": 0,
3682
3697
  "input_cost_per_token_above_128k_tokens": 0,
3683
3698
  "litellm_provider": "gemini",
@@ -3697,6 +3712,7 @@
3697
3712
  "tpm": 4000000
3698
3713
  },
3699
3714
  "gemini/gemini-1.5-pro-latest": {
3715
+ "deprecation_date": "2025-09-29",
3700
3716
  "input_cost_per_token": 3.5e-06,
3701
3717
  "input_cost_per_token_above_128k_tokens": 7e-06,
3702
3718
  "litellm_provider": "gemini",
@@ -4063,7 +4079,7 @@
4063
4079
  "litellm_provider": "openrouter",
4064
4080
  "max_input_tokens": 66000,
4065
4081
  "max_output_tokens": 4096,
4066
- "max_tokens": 8192,
4082
+ "max_tokens": 4096,
4067
4083
  "mode": "chat",
4068
4084
  "output_cost_per_token": 2.8e-07,
4069
4085
  "supports_prompt_caching": true,
@@ -4394,7 +4410,7 @@
4394
4410
  "input_cost_per_token": 1.63e-06,
4395
4411
  "litellm_provider": "openrouter",
4396
4412
  "max_output_tokens": 8191,
4397
- "max_tokens": 100000,
4413
+ "max_tokens": 8191,
4398
4414
  "mode": "chat",
4399
4415
  "output_cost_per_token": 5.51e-06,
4400
4416
  "supports_tool_choice": true
@@ -4403,7 +4419,7 @@
4403
4419
  "input_cost_per_token": 1.102e-05,
4404
4420
  "litellm_provider": "openrouter",
4405
4421
  "max_output_tokens": 8191,
4406
- "max_tokens": 100000,
4422
+ "max_tokens": 8191,
4407
4423
  "mode": "chat",
4408
4424
  "output_cost_per_token": 3.268e-05,
4409
4425
  "supports_tool_choice": true
@@ -5882,7 +5898,7 @@
5882
5898
  "litellm_provider": "bedrock",
5883
5899
  "max_input_tokens": 128000,
5884
5900
  "max_output_tokens": 2048,
5885
- "max_tokens": 128000,
5901
+ "max_tokens": 2048,
5886
5902
  "mode": "chat",
5887
5903
  "output_cost_per_token": 2.2e-07,
5888
5904
  "supports_function_calling": true,
@@ -5893,7 +5909,7 @@
5893
5909
  "litellm_provider": "bedrock",
5894
5910
  "max_input_tokens": 128000,
5895
5911
  "max_output_tokens": 2048,
5896
- "max_tokens": 128000,
5912
+ "max_tokens": 2048,
5897
5913
  "mode": "chat",
5898
5914
  "output_cost_per_token": 9.9e-07,
5899
5915
  "supports_function_calling": true,
@@ -5904,7 +5920,7 @@
5904
5920
  "litellm_provider": "bedrock",
5905
5921
  "max_input_tokens": 128000,
5906
5922
  "max_output_tokens": 4096,
5907
- "max_tokens": 128000,
5923
+ "max_tokens": 4096,
5908
5924
  "mode": "chat",
5909
5925
  "output_cost_per_token": 1.6e-05,
5910
5926
  "supports_function_calling": true,
@@ -5915,7 +5931,7 @@
5915
5931
  "litellm_provider": "bedrock",
5916
5932
  "max_input_tokens": 128000,
5917
5933
  "max_output_tokens": 4096,
5918
- "max_tokens": 128000,
5934
+ "max_tokens": 4096,
5919
5935
  "mode": "chat",
5920
5936
  "output_cost_per_token": 1e-07,
5921
5937
  "supports_function_calling": true,
@@ -5926,7 +5942,7 @@
5926
5942
  "litellm_provider": "bedrock",
5927
5943
  "max_input_tokens": 128000,
5928
5944
  "max_output_tokens": 4096,
5929
- "max_tokens": 128000,
5945
+ "max_tokens": 4096,
5930
5946
  "mode": "chat",
5931
5947
  "output_cost_per_token": 1e-07,
5932
5948
  "supports_function_calling": true,
@@ -5937,7 +5953,7 @@
5937
5953
  "litellm_provider": "bedrock",
5938
5954
  "max_input_tokens": 128000,
5939
5955
  "max_output_tokens": 4096,
5940
- "max_tokens": 128000,
5956
+ "max_tokens": 4096,
5941
5957
  "mode": "chat",
5942
5958
  "output_cost_per_token": 1.3e-07,
5943
5959
  "supports_function_calling": true,
@@ -5948,7 +5964,7 @@
5948
5964
  "litellm_provider": "bedrock",
5949
5965
  "max_input_tokens": 128000,
5950
5966
  "max_output_tokens": 4096,
5951
- "max_tokens": 128000,
5967
+ "max_tokens": 4096,
5952
5968
  "mode": "chat",
5953
5969
  "output_cost_per_token": 1.5e-07,
5954
5970
  "supports_function_calling": true,
@@ -5959,7 +5975,7 @@
5959
5975
  "litellm_provider": "bedrock",
5960
5976
  "max_input_tokens": 128000,
5961
5977
  "max_output_tokens": 4096,
5962
- "max_tokens": 128000,
5978
+ "max_tokens": 4096,
5963
5979
  "mode": "chat",
5964
5980
  "output_cost_per_token": 1.5e-07,
5965
5981
  "supports_function_calling": true,
@@ -5970,7 +5986,7 @@
5970
5986
  "litellm_provider": "bedrock",
5971
5987
  "max_input_tokens": 128000,
5972
5988
  "max_output_tokens": 4096,
5973
- "max_tokens": 128000,
5989
+ "max_tokens": 4096,
5974
5990
  "mode": "chat",
5975
5991
  "output_cost_per_token": 1.9e-07,
5976
5992
  "supports_function_calling": true,
@@ -5981,7 +5997,7 @@
5981
5997
  "litellm_provider": "bedrock",
5982
5998
  "max_input_tokens": 128000,
5983
5999
  "max_output_tokens": 4096,
5984
- "max_tokens": 128000,
6000
+ "max_tokens": 4096,
5985
6001
  "mode": "chat",
5986
6002
  "output_cost_per_token": 3.5e-07,
5987
6003
  "supports_function_calling": true,
@@ -5993,7 +6009,7 @@
5993
6009
  "litellm_provider": "bedrock",
5994
6010
  "max_input_tokens": 128000,
5995
6011
  "max_output_tokens": 4096,
5996
- "max_tokens": 128000,
6012
+ "max_tokens": 4096,
5997
6013
  "mode": "chat",
5998
6014
  "output_cost_per_token": 3.5e-07,
5999
6015
  "supports_function_calling": true,
@@ -6005,7 +6021,7 @@
6005
6021
  "litellm_provider": "bedrock",
6006
6022
  "max_input_tokens": 128000,
6007
6023
  "max_output_tokens": 4096,
6008
- "max_tokens": 128000,
6024
+ "max_tokens": 4096,
6009
6025
  "mode": "chat",
6010
6026
  "output_cost_per_token": 2e-06,
6011
6027
  "supports_function_calling": true,
@@ -6017,7 +6033,7 @@
6017
6033
  "litellm_provider": "bedrock",
6018
6034
  "max_input_tokens": 128000,
6019
6035
  "max_output_tokens": 4096,
6020
- "max_tokens": 128000,
6036
+ "max_tokens": 4096,
6021
6037
  "mode": "chat",
6022
6038
  "output_cost_per_token": 2e-06,
6023
6039
  "supports_function_calling": true,
@@ -6208,7 +6224,7 @@
6208
6224
  "litellm_provider": "ollama",
6209
6225
  "max_input_tokens": 32768,
6210
6226
  "max_output_tokens": 8192,
6211
- "max_tokens": 32768,
6227
+ "max_tokens": 8192,
6212
6228
  "mode": "chat",
6213
6229
  "output_cost_per_token": 0.0,
6214
6230
  "supports_function_calling": false
@@ -6218,7 +6234,7 @@
6218
6234
  "litellm_provider": "ollama",
6219
6235
  "max_input_tokens": 32768,
6220
6236
  "max_output_tokens": 8192,
6221
- "max_tokens": 32768,
6237
+ "max_tokens": 8192,
6222
6238
  "mode": "chat",
6223
6239
  "output_cost_per_token": 0.0,
6224
6240
  "supports_function_calling": true
@@ -6238,7 +6254,7 @@
6238
6254
  "litellm_provider": "ollama",
6239
6255
  "max_input_tokens": 32768,
6240
6256
  "max_output_tokens": 8192,
6241
- "max_tokens": 32768,
6257
+ "max_tokens": 8192,
6242
6258
  "mode": "chat",
6243
6259
  "output_cost_per_token": 0.0,
6244
6260
  "supports_function_calling": true
@@ -6258,7 +6274,7 @@
6258
6274
  "litellm_provider": "ollama",
6259
6275
  "max_input_tokens": 32768,
6260
6276
  "max_output_tokens": 8192,
6261
- "max_tokens": 32768,
6277
+ "max_tokens": 8192,
6262
6278
  "mode": "chat",
6263
6279
  "output_cost_per_token": 0.0,
6264
6280
  "supports_function_calling": true
@@ -6340,7 +6356,7 @@
6340
6356
  "litellm_provider": "ollama",
6341
6357
  "max_input_tokens": 8192,
6342
6358
  "max_output_tokens": 8192,
6343
- "max_tokens": 32768,
6359
+ "max_tokens": 8192,
6344
6360
  "mode": "chat",
6345
6361
  "output_cost_per_token": 0.0,
6346
6362
  "supports_function_calling": true
@@ -6350,7 +6366,7 @@
6350
6366
  "litellm_provider": "ollama",
6351
6367
  "max_input_tokens": 65536,
6352
6368
  "max_output_tokens": 8192,
6353
- "max_tokens": 65536,
6369
+ "max_tokens": 8192,
6354
6370
  "mode": "chat",
6355
6371
  "output_cost_per_token": 0.0,
6356
6372
  "supports_function_calling": true
@@ -7989,7 +8005,7 @@
7989
8005
  "litellm_provider": "bedrock",
7990
8006
  "max_input_tokens": 128000,
7991
8007
  "max_output_tokens": 2048,
7992
- "max_tokens": 128000,
8008
+ "max_tokens": 2048,
7993
8009
  "mode": "chat",
7994
8010
  "output_cost_per_token": 2.2e-07,
7995
8011
  "supports_function_calling": true,
@@ -8000,7 +8016,7 @@
8000
8016
  "litellm_provider": "bedrock",
8001
8017
  "max_input_tokens": 128000,
8002
8018
  "max_output_tokens": 2048,
8003
- "max_tokens": 128000,
8019
+ "max_tokens": 2048,
8004
8020
  "mode": "chat",
8005
8021
  "output_cost_per_token": 9.9e-07,
8006
8022
  "supports_function_calling": true,
@@ -8011,7 +8027,7 @@
8011
8027
  "litellm_provider": "bedrock",
8012
8028
  "max_input_tokens": 128000,
8013
8029
  "max_output_tokens": 4096,
8014
- "max_tokens": 128000,
8030
+ "max_tokens": 4096,
8015
8031
  "mode": "chat",
8016
8032
  "output_cost_per_token": 1.6e-05,
8017
8033
  "supports_function_calling": true,
@@ -8042,7 +8058,7 @@
8042
8058
  "litellm_provider": "openai",
8043
8059
  "max_input_tokens": 32768,
8044
8060
  "max_output_tokens": 0,
8045
- "max_tokens": 32768,
8061
+ "max_tokens": 0,
8046
8062
  "mode": "moderation",
8047
8063
  "output_cost_per_token": 0.0
8048
8064
  },
@@ -8051,7 +8067,7 @@
8051
8067
  "litellm_provider": "openai",
8052
8068
  "max_input_tokens": 32768,
8053
8069
  "max_output_tokens": 0,
8054
- "max_tokens": 32768,
8070
+ "max_tokens": 0,
8055
8071
  "mode": "moderation",
8056
8072
  "output_cost_per_token": 0.0
8057
8073
  },
@@ -8060,7 +8076,7 @@
8060
8076
  "litellm_provider": "openai",
8061
8077
  "max_input_tokens": 32768,
8062
8078
  "max_output_tokens": 0,
8063
- "max_tokens": 32768,
8079
+ "max_tokens": 0,
8064
8080
  "mode": "moderation",
8065
8081
  "output_cost_per_token": 0.0
8066
8082
  },
@@ -8510,6 +8526,7 @@
8510
8526
  "source": "https://cloud.google.com/vertex-ai/generative-ai/docs/learn/models"
8511
8527
  },
8512
8528
  "gemini/gemini-1.5-flash-8b": {
8529
+ "deprecation_date": "2025-09-29",
8513
8530
  "input_cost_per_token": 0,
8514
8531
  "input_cost_per_token_above_128k_tokens": 0,
8515
8532
  "litellm_provider": "gemini",
@@ -9176,6 +9193,7 @@
9176
9193
  "supports_web_search": true
9177
9194
  },
9178
9195
  "gemini-2.0-flash-thinking-exp": {
9196
+ "deprecation_date": "2025-12-02",
9179
9197
  "cache_read_input_token_cost": 0.0,
9180
9198
  "input_cost_per_audio_per_second": 0,
9181
9199
  "input_cost_per_audio_per_second_above_128k_tokens": 0,
@@ -9224,6 +9242,7 @@
9224
9242
  "supports_web_search": true
9225
9243
  },
9226
9244
  "gemini-2.0-flash-thinking-exp-01-21": {
9245
+ "deprecation_date": "2025-12-02",
9227
9246
  "cache_read_input_token_cost": 0.0,
9228
9247
  "input_cost_per_audio_per_second": 0,
9229
9248
  "input_cost_per_audio_per_second_above_128k_tokens": 0,
@@ -9311,6 +9330,7 @@
9311
9330
  "tpm": 10000000
9312
9331
  },
9313
9332
  "gemini/gemini-2.0-flash-lite-preview-02-05": {
9333
+ "deprecation_date": "2025-12-02",
9314
9334
  "cache_read_input_token_cost": 1.875e-08,
9315
9335
  "input_cost_per_audio_token": 7.5e-08,
9316
9336
  "input_cost_per_token": 7.5e-08,
@@ -9348,6 +9368,7 @@
9348
9368
  "tpm": 10000000
9349
9369
  },
9350
9370
  "gemini/gemini-2.0-flash-thinking-exp": {
9371
+ "deprecation_date": "2025-12-02",
9351
9372
  "cache_read_input_token_cost": 0.0,
9352
9373
  "input_cost_per_audio_per_second": 0,
9353
9374
  "input_cost_per_audio_per_second_above_128k_tokens": 0,
@@ -9366,7 +9387,7 @@
9366
9387
  "max_input_tokens": 1048576,
9367
9388
  "max_output_tokens": 65536,
9368
9389
  "max_pdf_size_mb": 30,
9369
- "max_tokens": 8192,
9390
+ "max_tokens": 65536,
9370
9391
  "max_video_length": 1,
9371
9392
  "max_videos_per_prompt": 10,
9372
9393
  "mode": "chat",
@@ -10401,6 +10422,7 @@
10401
10422
  "tpm": 1000000
10402
10423
  },
10403
10424
  "gemini/gemini-2.0-flash-thinking-exp-01-21": {
10425
+ "deprecation_date": "2025-12-02",
10404
10426
  "cache_read_input_token_cost": 0.0,
10405
10427
  "input_cost_per_audio_per_second": 0,
10406
10428
  "input_cost_per_audio_per_second_above_128k_tokens": 0,
@@ -10419,7 +10441,7 @@
10419
10441
  "max_input_tokens": 1048576,
10420
10442
  "max_output_tokens": 65536,
10421
10443
  "max_pdf_size_mb": 30,
10422
- "max_tokens": 8192,
10444
+ "max_tokens": 65536,
10423
10445
  "max_video_length": 1,
10424
10446
  "max_videos_per_prompt": 10,
10425
10447
  "mode": "chat",
@@ -10507,6 +10529,7 @@
10507
10529
  "supports_vision": true
10508
10530
  },
10509
10531
  "vertex_ai/imagen-3.0-generate-002": {
10532
+ "deprecation_date": "2025-11-10",
10510
10533
  "litellm_provider": "vertex_ai-image-models",
10511
10534
  "mode": "image_generation",
10512
10535
  "output_cost_per_image": 0.04,
@@ -10595,7 +10618,7 @@
10595
10618
  "litellm_provider": "snowflake",
10596
10619
  "max_input_tokens": 32768,
10597
10620
  "max_output_tokens": 8192,
10598
- "max_tokens": 32768,
10621
+ "max_tokens": 8192,
10599
10622
  "mode": "chat",
10600
10623
  "supports_reasoning": true
10601
10624
  },
@@ -10603,14 +10626,14 @@
10603
10626
  "litellm_provider": "snowflake",
10604
10627
  "max_input_tokens": 4096,
10605
10628
  "max_output_tokens": 8192,
10606
- "max_tokens": 4096,
10629
+ "max_tokens": 8192,
10607
10630
  "mode": "chat"
10608
10631
  },
10609
10632
  "snowflake/claude-3-5-sonnet": {
10610
10633
  "litellm_provider": "snowflake",
10611
10634
  "max_input_tokens": 18000,
10612
10635
  "max_output_tokens": 8192,
10613
- "max_tokens": 18000,
10636
+ "max_tokens": 8192,
10614
10637
  "mode": "chat",
10615
10638
  "supports_computer_use": true
10616
10639
  },
@@ -10618,147 +10641,147 @@
10618
10641
  "litellm_provider": "snowflake",
10619
10642
  "max_input_tokens": 32000,
10620
10643
  "max_output_tokens": 8192,
10621
- "max_tokens": 32000,
10644
+ "max_tokens": 8192,
10622
10645
  "mode": "chat"
10623
10646
  },
10624
10647
  "snowflake/mistral-large2": {
10625
10648
  "litellm_provider": "snowflake",
10626
10649
  "max_input_tokens": 128000,
10627
10650
  "max_output_tokens": 8192,
10628
- "max_tokens": 128000,
10651
+ "max_tokens": 8192,
10629
10652
  "mode": "chat"
10630
10653
  },
10631
10654
  "snowflake/reka-flash": {
10632
10655
  "litellm_provider": "snowflake",
10633
10656
  "max_input_tokens": 100000,
10634
10657
  "max_output_tokens": 8192,
10635
- "max_tokens": 100000,
10658
+ "max_tokens": 8192,
10636
10659
  "mode": "chat"
10637
10660
  },
10638
10661
  "snowflake/reka-core": {
10639
10662
  "litellm_provider": "snowflake",
10640
10663
  "max_input_tokens": 32000,
10641
10664
  "max_output_tokens": 8192,
10642
- "max_tokens": 32000,
10665
+ "max_tokens": 8192,
10643
10666
  "mode": "chat"
10644
10667
  },
10645
10668
  "snowflake/jamba-instruct": {
10646
10669
  "litellm_provider": "snowflake",
10647
10670
  "max_input_tokens": 256000,
10648
10671
  "max_output_tokens": 8192,
10649
- "max_tokens": 256000,
10672
+ "max_tokens": 8192,
10650
10673
  "mode": "chat"
10651
10674
  },
10652
10675
  "snowflake/jamba-1.5-mini": {
10653
10676
  "litellm_provider": "snowflake",
10654
10677
  "max_input_tokens": 256000,
10655
10678
  "max_output_tokens": 8192,
10656
- "max_tokens": 256000,
10679
+ "max_tokens": 8192,
10657
10680
  "mode": "chat"
10658
10681
  },
10659
10682
  "snowflake/jamba-1.5-large": {
10660
10683
  "litellm_provider": "snowflake",
10661
10684
  "max_input_tokens": 256000,
10662
10685
  "max_output_tokens": 8192,
10663
- "max_tokens": 256000,
10686
+ "max_tokens": 8192,
10664
10687
  "mode": "chat"
10665
10688
  },
10666
10689
  "snowflake/mixtral-8x7b": {
10667
10690
  "litellm_provider": "snowflake",
10668
10691
  "max_input_tokens": 32000,
10669
10692
  "max_output_tokens": 8192,
10670
- "max_tokens": 32000,
10693
+ "max_tokens": 8192,
10671
10694
  "mode": "chat"
10672
10695
  },
10673
10696
  "snowflake/llama2-70b-chat": {
10674
10697
  "litellm_provider": "snowflake",
10675
10698
  "max_input_tokens": 4096,
10676
10699
  "max_output_tokens": 8192,
10677
- "max_tokens": 4096,
10700
+ "max_tokens": 8192,
10678
10701
  "mode": "chat"
10679
10702
  },
10680
10703
  "snowflake/llama3-8b": {
10681
10704
  "litellm_provider": "snowflake",
10682
10705
  "max_input_tokens": 8000,
10683
10706
  "max_output_tokens": 8192,
10684
- "max_tokens": 8000,
10707
+ "max_tokens": 8192,
10685
10708
  "mode": "chat"
10686
10709
  },
10687
10710
  "snowflake/llama3-70b": {
10688
10711
  "litellm_provider": "snowflake",
10689
10712
  "max_input_tokens": 8000,
10690
10713
  "max_output_tokens": 8192,
10691
- "max_tokens": 8000,
10714
+ "max_tokens": 8192,
10692
10715
  "mode": "chat"
10693
10716
  },
10694
10717
  "snowflake/llama3.1-8b": {
10695
10718
  "litellm_provider": "snowflake",
10696
10719
  "max_input_tokens": 128000,
10697
10720
  "max_output_tokens": 8192,
10698
- "max_tokens": 128000,
10721
+ "max_tokens": 8192,
10699
10722
  "mode": "chat"
10700
10723
  },
10701
10724
  "snowflake/llama3.1-70b": {
10702
10725
  "litellm_provider": "snowflake",
10703
10726
  "max_input_tokens": 128000,
10704
10727
  "max_output_tokens": 8192,
10705
- "max_tokens": 128000,
10728
+ "max_tokens": 8192,
10706
10729
  "mode": "chat"
10707
10730
  },
10708
10731
  "snowflake/llama3.3-70b": {
10709
10732
  "litellm_provider": "snowflake",
10710
10733
  "max_input_tokens": 128000,
10711
10734
  "max_output_tokens": 8192,
10712
- "max_tokens": 128000,
10735
+ "max_tokens": 8192,
10713
10736
  "mode": "chat"
10714
10737
  },
10715
10738
  "snowflake/snowflake-llama-3.3-70b": {
10716
10739
  "litellm_provider": "snowflake",
10717
10740
  "max_input_tokens": 8000,
10718
10741
  "max_output_tokens": 8192,
10719
- "max_tokens": 8000,
10742
+ "max_tokens": 8192,
10720
10743
  "mode": "chat"
10721
10744
  },
10722
10745
  "snowflake/llama3.1-405b": {
10723
10746
  "litellm_provider": "snowflake",
10724
10747
  "max_input_tokens": 128000,
10725
10748
  "max_output_tokens": 8192,
10726
- "max_tokens": 128000,
10749
+ "max_tokens": 8192,
10727
10750
  "mode": "chat"
10728
10751
  },
10729
10752
  "snowflake/snowflake-llama-3.1-405b": {
10730
10753
  "litellm_provider": "snowflake",
10731
10754
  "max_input_tokens": 8000,
10732
10755
  "max_output_tokens": 8192,
10733
- "max_tokens": 8000,
10756
+ "max_tokens": 8192,
10734
10757
  "mode": "chat"
10735
10758
  },
10736
10759
  "snowflake/llama3.2-1b": {
10737
10760
  "litellm_provider": "snowflake",
10738
10761
  "max_input_tokens": 128000,
10739
10762
  "max_output_tokens": 8192,
10740
- "max_tokens": 128000,
10763
+ "max_tokens": 8192,
10741
10764
  "mode": "chat"
10742
10765
  },
10743
10766
  "snowflake/llama3.2-3b": {
10744
10767
  "litellm_provider": "snowflake",
10745
10768
  "max_input_tokens": 128000,
10746
10769
  "max_output_tokens": 8192,
10747
- "max_tokens": 128000,
10770
+ "max_tokens": 8192,
10748
10771
  "mode": "chat"
10749
10772
  },
10750
10773
  "snowflake/mistral-7b": {
10751
10774
  "litellm_provider": "snowflake",
10752
10775
  "max_input_tokens": 32000,
10753
10776
  "max_output_tokens": 8192,
10754
- "max_tokens": 32000,
10777
+ "max_tokens": 8192,
10755
10778
  "mode": "chat"
10756
10779
  },
10757
10780
  "snowflake/gemma-7b": {
10758
10781
  "litellm_provider": "snowflake",
10759
10782
  "max_input_tokens": 8000,
10760
10783
  "max_output_tokens": 8192,
10761
- "max_tokens": 8000,
10784
+ "max_tokens": 8192,
10762
10785
  "mode": "chat"
10763
10786
  },
10764
10787
  "azure/global/gpt-4o-2024-11-20": {
@@ -11211,7 +11234,7 @@
11211
11234
  "litellm_provider": "watsonx",
11212
11235
  "max_input_tokens": 8192,
11213
11236
  "max_output_tokens": 1024,
11214
- "max_tokens": 8192,
11237
+ "max_tokens": 1024,
11215
11238
  "mode": "chat",
11216
11239
  "output_cost_per_token": 2e-07,
11217
11240
  "supports_audio_input": false,
@@ -11366,16 +11389,16 @@
11366
11389
  "supports_vision": true
11367
11390
  },
11368
11391
  "gpt-image-1": {
11369
- "input_cost_per_image": 0.042,
11370
- "input_cost_per_pixel": 4.0054321e-08,
11371
- "input_cost_per_token": 5e-06,
11392
+ "cache_read_input_image_token_cost": 2.5e-06,
11393
+ "cache_read_input_token_cost": 1.25e-06,
11372
11394
  "input_cost_per_image_token": 1e-05,
11395
+ "input_cost_per_token": 5e-06,
11373
11396
  "litellm_provider": "openai",
11374
11397
  "mode": "image_generation",
11375
- "output_cost_per_pixel": 0.0,
11376
- "output_cost_per_token": 4e-05,
11398
+ "output_cost_per_image_token": 4e-05,
11377
11399
  "supported_endpoints": [
11378
- "/v1/images/generations"
11400
+ "/v1/images/generations",
11401
+ "/v1/images/edits"
11379
11402
  ]
11380
11403
  },
11381
11404
  "low/1024-x-1024/gpt-image-1": {
@@ -11984,12 +12007,16 @@
11984
12007
  "supports_vision": true
11985
12008
  },
11986
12009
  "azure/gpt-image-1": {
11987
- "input_cost_per_pixel": 4.0054321e-08,
12010
+ "cache_read_input_image_token_cost": 2.5e-06,
12011
+ "cache_read_input_token_cost": 1.25e-06,
12012
+ "input_cost_per_image_token": 1e-05,
12013
+ "input_cost_per_token": 5e-06,
11988
12014
  "litellm_provider": "azure",
11989
12015
  "mode": "image_generation",
11990
- "output_cost_per_pixel": 0.0,
12016
+ "output_cost_per_image_token": 4e-05,
11991
12017
  "supported_endpoints": [
11992
- "/v1/images/generations"
12018
+ "/v1/images/generations",
12019
+ "/v1/images/edits"
11993
12020
  ]
11994
12021
  },
11995
12022
  "azure/low/1024-x-1024/gpt-image-1": {
@@ -12194,7 +12221,7 @@
12194
12221
  "litellm_provider": "meta_llama",
12195
12222
  "max_input_tokens": 10000000,
12196
12223
  "max_output_tokens": 4028,
12197
- "max_tokens": 128000,
12224
+ "max_tokens": 4028,
12198
12225
  "mode": "chat",
12199
12226
  "source": "https://llama.developer.meta.com/docs/models",
12200
12227
  "supported_modalities": [
@@ -12211,7 +12238,7 @@
12211
12238
  "litellm_provider": "meta_llama",
12212
12239
  "max_input_tokens": 1000000,
12213
12240
  "max_output_tokens": 4028,
12214
- "max_tokens": 128000,
12241
+ "max_tokens": 4028,
12215
12242
  "mode": "chat",
12216
12243
  "source": "https://llama.developer.meta.com/docs/models",
12217
12244
  "supported_modalities": [
@@ -12228,7 +12255,7 @@
12228
12255
  "litellm_provider": "meta_llama",
12229
12256
  "max_input_tokens": 128000,
12230
12257
  "max_output_tokens": 4028,
12231
- "max_tokens": 128000,
12258
+ "max_tokens": 4028,
12232
12259
  "mode": "chat",
12233
12260
  "source": "https://llama.developer.meta.com/docs/models",
12234
12261
  "supported_modalities": [
@@ -12244,7 +12271,7 @@
12244
12271
  "litellm_provider": "meta_llama",
12245
12272
  "max_input_tokens": 128000,
12246
12273
  "max_output_tokens": 4028,
12247
- "max_tokens": 128000,
12274
+ "max_tokens": 4028,
12248
12275
  "mode": "chat",
12249
12276
  "source": "https://llama.developer.meta.com/docs/models",
12250
12277
  "supported_modalities": [
@@ -12542,6 +12569,7 @@
12542
12569
  "supports_web_search": true
12543
12570
  },
12544
12571
  "gemini-2.5-pro-preview-05-06": {
12572
+ "deprecation_date": "2025-12-02",
12545
12573
  "cache_read_input_token_cost": 3.125e-07,
12546
12574
  "input_cost_per_audio_token": 1.25e-06,
12547
12575
  "input_cost_per_token": 1.25e-06,
@@ -12590,6 +12618,7 @@
12590
12618
  "supports_web_search": true
12591
12619
  },
12592
12620
  "gemini-2.5-pro-preview-03-25": {
12621
+ "deprecation_date": "2025-12-02",
12593
12622
  "cache_read_input_token_cost": 3.125e-07,
12594
12623
  "input_cost_per_audio_token": 1.25e-06,
12595
12624
  "input_cost_per_token": 1.25e-06,
@@ -12671,6 +12700,7 @@
12671
12700
  "tpm": 4000000
12672
12701
  },
12673
12702
  "gemini/gemini-2.5-pro-preview-05-06": {
12703
+ "deprecation_date": "2025-12-02",
12674
12704
  "cache_read_input_token_cost": 3.125e-07,
12675
12705
  "input_cost_per_audio_token": 7e-07,
12676
12706
  "input_cost_per_token": 1.25e-06,
@@ -12712,6 +12742,7 @@
12712
12742
  "tpm": 10000000
12713
12743
  },
12714
12744
  "gemini/gemini-2.5-pro-preview-03-25": {
12745
+ "deprecation_date": "2025-12-02",
12715
12746
  "cache_read_input_token_cost": 3.125e-07,
12716
12747
  "input_cost_per_audio_token": 7e-07,
12717
12748
  "input_cost_per_token": 1.25e-06,
@@ -13208,7 +13239,7 @@
13208
13239
  "litellm_provider": "databricks",
13209
13240
  "max_input_tokens": 200000,
13210
13241
  "max_output_tokens": 128000,
13211
- "max_tokens": 200000,
13242
+ "max_tokens": 128000,
13212
13243
  "metadata": {
13213
13244
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
13214
13245
  },
@@ -13584,14 +13615,14 @@
13584
13615
  "litellm_provider": "featherless_ai",
13585
13616
  "max_input_tokens": 32768,
13586
13617
  "max_output_tokens": 4096,
13587
- "max_tokens": 32768,
13618
+ "max_tokens": 4096,
13588
13619
  "mode": "chat"
13589
13620
  },
13590
13621
  "featherless_ai/featherless-ai/Qwerky-QwQ-32B": {
13591
13622
  "litellm_provider": "featherless_ai",
13592
13623
  "max_input_tokens": 32768,
13593
13624
  "max_output_tokens": 4096,
13594
- "max_tokens": 32768,
13625
+ "max_tokens": 4096,
13595
13626
  "mode": "chat"
13596
13627
  },
13597
13628
  "sambanova/Llama-4-Maverick-17B-128E-Instruct": {
@@ -13759,6 +13790,7 @@
13759
13790
  "tpm": 250000
13760
13791
  },
13761
13792
  "gemini/gemini-2.5-flash-preview-05-20": {
13793
+ "deprecation_date": "2025-11-18",
13762
13794
  "cache_read_input_token_cost": 7.5e-08,
13763
13795
  "input_cost_per_audio_token": 1e-06,
13764
13796
  "input_cost_per_token": 3e-07,
@@ -13804,6 +13836,7 @@
13804
13836
  "tpm": 250000
13805
13837
  },
13806
13838
  "gemini-2.5-flash-preview-05-20": {
13839
+ "deprecation_date": "2025-11-18",
13807
13840
  "cache_read_input_token_cost": 7.5e-08,
13808
13841
  "input_cost_per_audio_token": 1e-06,
13809
13842
  "input_cost_per_token": 3e-07,
@@ -13849,6 +13882,7 @@
13849
13882
  "supports_web_search": true
13850
13883
  },
13851
13884
  "gemini-2.0-flash-preview-image-generation": {
13885
+ "deprecation_date": "2025-11-14",
13852
13886
  "cache_read_input_token_cost": 2.5e-08,
13853
13887
  "input_cost_per_audio_token": 7e-07,
13854
13888
  "input_cost_per_token": 1e-07,
@@ -13922,6 +13956,7 @@
13922
13956
  "supports_web_search": true
13923
13957
  },
13924
13958
  "gemini/gemini-2.0-flash-preview-image-generation": {
13959
+ "deprecation_date": "2025-11-14",
13925
13960
  "cache_read_input_token_cost": 2.5e-08,
13926
13961
  "input_cost_per_audio_token": 7e-07,
13927
13962
  "input_cost_per_token": 1e-07,
@@ -14424,7 +14459,7 @@
14424
14459
  "litellm_provider": "anthropic",
14425
14460
  "max_input_tokens": 1000000,
14426
14461
  "max_output_tokens": 64000,
14427
- "max_tokens": 1000000,
14462
+ "max_tokens": 64000,
14428
14463
  "mode": "chat",
14429
14464
  "output_cost_per_token": 1.5e-05,
14430
14465
  "output_cost_per_token_above_200k_tokens": 2.25e-05,
@@ -15496,6 +15531,7 @@
15496
15531
  "supports_web_search": true
15497
15532
  },
15498
15533
  "gemini-2.5-flash-lite-preview-06-17": {
15534
+ "deprecation_date": "2025-11-18",
15499
15535
  "cache_read_input_token_cost": 2.5e-08,
15500
15536
  "input_cost_per_audio_token": 5e-07,
15501
15537
  "input_cost_per_token": 1e-07,
@@ -15671,6 +15707,7 @@
15671
15707
  "tpm": 8000000
15672
15708
  },
15673
15709
  "gemini/gemini-2.5-flash-lite-preview-06-17": {
15710
+ "deprecation_date": "2025-11-18",
15674
15711
  "cache_read_input_token_cost": 2.5e-08,
15675
15712
  "input_cost_per_audio_token": 5e-07,
15676
15713
  "input_cost_per_token": 1e-07,
@@ -16206,7 +16243,7 @@
16206
16243
  "litellm_provider": "watsonx",
16207
16244
  "max_input_tokens": 131072,
16208
16245
  "max_output_tokens": 16384,
16209
- "max_tokens": 131072,
16246
+ "max_tokens": 16384,
16210
16247
  "mode": "chat",
16211
16248
  "output_cost_per_token": 1e-05,
16212
16249
  "supports_audio_input": false,
@@ -16327,7 +16364,7 @@
16327
16364
  "litellm_provider": "bedrock_converse",
16328
16365
  "max_input_tokens": 128000,
16329
16366
  "max_output_tokens": 4096,
16330
- "max_tokens": 128000,
16367
+ "max_tokens": 4096,
16331
16368
  "mode": "chat",
16332
16369
  "output_cost_per_token": 6e-06,
16333
16370
  "supports_function_calling": true,
@@ -16338,7 +16375,7 @@
16338
16375
  "litellm_provider": "bedrock_converse",
16339
16376
  "max_input_tokens": 128000,
16340
16377
  "max_output_tokens": 4096,
16341
- "max_tokens": 128000,
16378
+ "max_tokens": 4096,
16342
16379
  "mode": "chat",
16343
16380
  "output_cost_per_token": 6e-06,
16344
16381
  "supports_function_calling": true,
@@ -16349,7 +16386,7 @@
16349
16386
  "litellm_provider": "dashscope",
16350
16387
  "max_input_tokens": 30720,
16351
16388
  "max_output_tokens": 8192,
16352
- "max_tokens": 32768,
16389
+ "max_tokens": 8192,
16353
16390
  "mode": "chat",
16354
16391
  "output_cost_per_token": 6.4e-06,
16355
16392
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -16361,7 +16398,7 @@
16361
16398
  "litellm_provider": "dashscope",
16362
16399
  "max_input_tokens": 997952,
16363
16400
  "max_output_tokens": 32768,
16364
- "max_tokens": 1000000,
16401
+ "max_tokens": 32768,
16365
16402
  "mode": "chat",
16366
16403
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
16367
16404
  "supports_function_calling": true,
@@ -16393,7 +16430,7 @@
16393
16430
  "litellm_provider": "dashscope",
16394
16431
  "max_input_tokens": 1000000,
16395
16432
  "max_output_tokens": 16384,
16396
- "max_tokens": 1000000,
16433
+ "max_tokens": 16384,
16397
16434
  "mode": "chat",
16398
16435
  "output_cost_per_reasoning_token": 5e-07,
16399
16436
  "output_cost_per_token": 2e-07,
@@ -16406,7 +16443,7 @@
16406
16443
  "litellm_provider": "dashscope",
16407
16444
  "max_input_tokens": 129024,
16408
16445
  "max_output_tokens": 16384,
16409
- "max_tokens": 131072,
16446
+ "max_tokens": 16384,
16410
16447
  "mode": "chat",
16411
16448
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
16412
16449
  "supports_function_calling": true,
@@ -17039,7 +17076,7 @@
17039
17076
  "litellm_provider": "openrouter",
17040
17077
  "max_input_tokens": 8192,
17041
17078
  "max_output_tokens": 2048,
17042
- "max_tokens": 8192,
17079
+ "max_tokens": 2048,
17043
17080
  "mode": "chat",
17044
17081
  "output_cost_per_token": 6.3e-07,
17045
17082
  "supports_tool_choice": true,
@@ -17050,7 +17087,7 @@
17050
17087
  "litellm_provider": "fireworks_ai",
17051
17088
  "max_input_tokens": 131072,
17052
17089
  "max_output_tokens": 16384,
17053
- "max_tokens": 131072,
17090
+ "max_tokens": 16384,
17054
17091
  "mode": "chat",
17055
17092
  "output_cost_per_token": 2.5e-06,
17056
17093
  "source": "https://fireworks.ai/models/fireworks/kimi-k2-instruct",
@@ -17183,7 +17220,7 @@
17183
17220
  "litellm_provider": "lambda_ai",
17184
17221
  "max_input_tokens": 131072,
17185
17222
  "max_output_tokens": 8192,
17186
- "max_tokens": 131072,
17223
+ "max_tokens": 8192,
17187
17224
  "mode": "chat",
17188
17225
  "output_cost_per_token": 1e-07,
17189
17226
  "supports_function_calling": true,
@@ -17196,7 +17233,7 @@
17196
17233
  "litellm_provider": "lambda_ai",
17197
17234
  "max_input_tokens": 16384,
17198
17235
  "max_output_tokens": 8192,
17199
- "max_tokens": 16384,
17236
+ "max_tokens": 8192,
17200
17237
  "mode": "chat",
17201
17238
  "output_cost_per_token": 1e-07,
17202
17239
  "supports_function_calling": true,
@@ -17370,6 +17407,7 @@
17370
17407
  "supports_vision": false
17371
17408
  },
17372
17409
  "gemini/gemini-2.0-flash-live-001": {
17410
+ "deprecation_date": "2025-12-09",
17373
17411
  "cache_read_input_token_cost": 7.5e-08,
17374
17412
  "input_cost_per_audio_token": 2.1e-06,
17375
17413
  "input_cost_per_image": 2.1e-06,
@@ -17470,7 +17508,7 @@
17470
17508
  "litellm_provider": "vertex_ai-llama_models",
17471
17509
  "max_input_tokens": 128000,
17472
17510
  "max_output_tokens": 2048,
17473
- "max_tokens": 128000,
17511
+ "max_tokens": 2048,
17474
17512
  "metadata": {
17475
17513
  "notes": "VertexAI states that The Llama 3.1 API service for llama-3.1-70b-instruct-maas and llama-3.1-8b-instruct-maas are in public preview and at no cost."
17476
17514
  },
@@ -17486,7 +17524,7 @@
17486
17524
  "litellm_provider": "vertex_ai-llama_models",
17487
17525
  "max_input_tokens": 128000,
17488
17526
  "max_output_tokens": 2048,
17489
- "max_tokens": 128000,
17527
+ "max_tokens": 2048,
17490
17528
  "mode": "chat",
17491
17529
  "output_cost_per_token": 0.0,
17492
17530
  "source": "https://console.cloud.google.com/vertex-ai/publishers/meta/model-garden/llama-3.2-90b-vision-instruct-maas",
@@ -17499,7 +17537,7 @@
17499
17537
  "litellm_provider": "vertex_ai-llama_models",
17500
17538
  "max_input_tokens": 128000,
17501
17539
  "max_output_tokens": 2048,
17502
- "max_tokens": 128000,
17540
+ "max_tokens": 2048,
17503
17541
  "mode": "chat",
17504
17542
  "output_cost_per_token": 1.6e-05,
17505
17543
  "source": "https://console.cloud.google.com/vertex-ai/publishers/meta/model-garden/llama-3.2-90b-vision-instruct-maas",
@@ -17867,6 +17905,7 @@
17867
17905
  "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing"
17868
17906
  },
17869
17907
  "gemini/imagen-3.0-generate-002": {
17908
+ "deprecation_date": "2025-11-10",
17870
17909
  "litellm_provider": "gemini",
17871
17910
  "mode": "image_generation",
17872
17911
  "output_cost_per_image": 0.04,
@@ -18730,7 +18769,7 @@
18730
18769
  "litellm_provider": "oci",
18731
18770
  "max_input_tokens": 512000,
18732
18771
  "max_output_tokens": 4000,
18733
- "max_tokens": 512000,
18772
+ "max_tokens": 4000,
18734
18773
  "mode": "chat",
18735
18774
  "output_cost_per_token": 7.2e-07,
18736
18775
  "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing",
@@ -18742,7 +18781,7 @@
18742
18781
  "litellm_provider": "oci",
18743
18782
  "max_input_tokens": 192000,
18744
18783
  "max_output_tokens": 4000,
18745
- "max_tokens": 192000,
18784
+ "max_tokens": 4000,
18746
18785
  "mode": "chat",
18747
18786
  "output_cost_per_token": 7.2e-07,
18748
18787
  "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing",
@@ -18754,7 +18793,7 @@
18754
18793
  "litellm_provider": "oci",
18755
18794
  "max_input_tokens": 128000,
18756
18795
  "max_output_tokens": 4000,
18757
- "max_tokens": 128000,
18796
+ "max_tokens": 4000,
18758
18797
  "mode": "chat",
18759
18798
  "output_cost_per_token": 7.2e-07,
18760
18799
  "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing",
@@ -18766,7 +18805,7 @@
18766
18805
  "litellm_provider": "oci",
18767
18806
  "max_input_tokens": 128000,
18768
18807
  "max_output_tokens": 4000,
18769
- "max_tokens": 128000,
18808
+ "max_tokens": 4000,
18770
18809
  "mode": "chat",
18771
18810
  "output_cost_per_token": 2e-06,
18772
18811
  "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing",
@@ -18778,7 +18817,7 @@
18778
18817
  "litellm_provider": "oci",
18779
18818
  "max_input_tokens": 128000,
18780
18819
  "max_output_tokens": 4000,
18781
- "max_tokens": 128000,
18820
+ "max_tokens": 4000,
18782
18821
  "mode": "chat",
18783
18822
  "output_cost_per_token": 1.068e-05,
18784
18823
  "source": "https://www.oracle.com/artificial-intelligence/generative-ai/generative-ai-service/pricing",
@@ -18876,13 +18915,13 @@
18876
18915
  "supports_tool_choice": true
18877
18916
  },
18878
18917
  "openrouter/openai/gpt-oss-20b": {
18879
- "input_cost_per_token": 1.8e-07,
18918
+ "input_cost_per_token": 2e-08,
18880
18919
  "litellm_provider": "openrouter",
18881
18920
  "max_input_tokens": 131072,
18882
18921
  "max_output_tokens": 32768,
18883
18922
  "max_tokens": 32768,
18884
18923
  "mode": "chat",
18885
- "output_cost_per_token": 8e-07,
18924
+ "output_cost_per_token": 1e-07,
18886
18925
  "source": "https://openrouter.ai/openai/gpt-oss-20b",
18887
18926
  "supports_function_calling": true,
18888
18927
  "supports_parallel_function_calling": true,
@@ -20292,7 +20331,7 @@
20292
20331
  "litellm_provider": "openrouter",
20293
20332
  "max_input_tokens": 163840,
20294
20333
  "max_output_tokens": 163840,
20295
- "max_tokens": 8192,
20334
+ "max_tokens": 163840,
20296
20335
  "mode": "chat",
20297
20336
  "output_cost_per_token": 8e-07,
20298
20337
  "supports_assistant_prefill": true,
@@ -20414,6 +20453,7 @@
20414
20453
  ]
20415
20454
  },
20416
20455
  "gemini/gemini-2.5-flash-image-preview": {
20456
+ "deprecation_date": "2026-01-15",
20417
20457
  "cache_read_input_token_cost": 7.5e-08,
20418
20458
  "input_cost_per_audio_token": 1e-06,
20419
20459
  "input_cost_per_token": 3e-07,
@@ -20463,8 +20503,10 @@
20463
20503
  "tpm": 8000000
20464
20504
  },
20465
20505
  "gemini-2.5-flash-image-preview": {
20506
+ "deprecation_date": "2026-01-15",
20466
20507
  "cache_read_input_token_cost": 7.5e-08,
20467
20508
  "input_cost_per_audio_token": 1e-06,
20509
+ "input_cost_per_image_token": 3e-07,
20468
20510
  "input_cost_per_token": 3e-07,
20469
20511
  "litellm_provider": "vertex_ai-language-models",
20470
20512
  "max_audio_length_hours": 8.4,
@@ -20733,7 +20775,7 @@
20733
20775
  "litellm_provider": "vercel_ai_gateway",
20734
20776
  "max_input_tokens": 262144,
20735
20777
  "max_output_tokens": 66536,
20736
- "max_tokens": 262144,
20778
+ "max_tokens": 66536,
20737
20779
  "mode": "chat",
20738
20780
  "output_cost_per_token": 1.6e-06
20739
20781
  },
@@ -20751,7 +20793,7 @@
20751
20793
  "litellm_provider": "vercel_ai_gateway",
20752
20794
  "max_input_tokens": 1048576,
20753
20795
  "max_output_tokens": 65536,
20754
- "max_tokens": 1048576,
20796
+ "max_tokens": 65536,
20755
20797
  "mode": "chat",
20756
20798
  "output_cost_per_token": 1e-05
20757
20799
  },
@@ -20760,7 +20802,7 @@
20760
20802
  "litellm_provider": "vercel_ai_gateway",
20761
20803
  "max_input_tokens": 128000,
20762
20804
  "max_output_tokens": 8192,
20763
- "max_tokens": 128000,
20805
+ "max_tokens": 8192,
20764
20806
  "mode": "chat",
20765
20807
  "output_cost_per_token": 9e-07
20766
20808
  },
@@ -20769,7 +20811,7 @@
20769
20811
  "litellm_provider": "vercel_ai_gateway",
20770
20812
  "max_input_tokens": 300000,
20771
20813
  "max_output_tokens": 8192,
20772
- "max_tokens": 300000,
20814
+ "max_tokens": 8192,
20773
20815
  "mode": "chat",
20774
20816
  "output_cost_per_token": 2.4e-07
20775
20817
  },
@@ -20778,7 +20820,7 @@
20778
20820
  "litellm_provider": "vercel_ai_gateway",
20779
20821
  "max_input_tokens": 131072,
20780
20822
  "max_output_tokens": 8192,
20781
- "max_tokens": 131072,
20823
+ "max_tokens": 8192,
20782
20824
  "mode": "chat",
20783
20825
  "output_cost_per_token": 3e-07
20784
20826
  },
@@ -20787,7 +20829,7 @@
20787
20829
  "litellm_provider": "vercel_ai_gateway",
20788
20830
  "max_input_tokens": 128000,
20789
20831
  "max_output_tokens": 8192,
20790
- "max_tokens": 128000,
20832
+ "max_tokens": 8192,
20791
20833
  "mode": "chat",
20792
20834
  "output_cost_per_token": 1e-07
20793
20835
  },
@@ -20796,7 +20838,7 @@
20796
20838
  "litellm_provider": "vercel_ai_gateway",
20797
20839
  "max_input_tokens": 32000,
20798
20840
  "max_output_tokens": 4000,
20799
- "max_tokens": 32000,
20841
+ "max_tokens": 4000,
20800
20842
  "mode": "chat",
20801
20843
  "output_cost_per_token": 3e-07
20802
20844
  },
@@ -20805,7 +20847,7 @@
20805
20847
  "litellm_provider": "vercel_ai_gateway",
20806
20848
  "max_input_tokens": 1000000,
20807
20849
  "max_output_tokens": 65536,
20808
- "max_tokens": 1000000,
20850
+ "max_tokens": 65536,
20809
20851
  "mode": "chat",
20810
20852
  "output_cost_per_token": 2.5e-06
20811
20853
  },
@@ -20814,7 +20856,7 @@
20814
20856
  "litellm_provider": "vercel_ai_gateway",
20815
20857
  "max_input_tokens": 32000,
20816
20858
  "max_output_tokens": 16384,
20817
- "max_tokens": 32000,
20859
+ "max_tokens": 16384,
20818
20860
  "mode": "chat",
20819
20861
  "output_cost_per_token": 1e-06
20820
20862
  },
@@ -20841,7 +20883,7 @@
20841
20883
  "litellm_provider": "vercel_ai_gateway",
20842
20884
  "max_input_tokens": 131072,
20843
20885
  "max_output_tokens": 4000,
20844
- "max_tokens": 131072,
20886
+ "max_tokens": 4000,
20845
20887
  "mode": "chat",
20846
20888
  "output_cost_per_token": 1e-05
20847
20889
  },
@@ -20859,7 +20901,7 @@
20859
20901
  "litellm_provider": "vercel_ai_gateway",
20860
20902
  "max_input_tokens": 128000,
20861
20903
  "max_output_tokens": 8192,
20862
- "max_tokens": 128000,
20904
+ "max_tokens": 8192,
20863
20905
  "mode": "chat",
20864
20906
  "output_cost_per_token": 7.2e-07
20865
20907
  },
@@ -20877,7 +20919,7 @@
20877
20919
  "litellm_provider": "vercel_ai_gateway",
20878
20920
  "max_input_tokens": 40960,
20879
20921
  "max_output_tokens": 16384,
20880
- "max_tokens": 40960,
20922
+ "max_tokens": 16384,
20881
20923
  "mode": "chat",
20882
20924
  "output_cost_per_token": 6e-07
20883
20925
  },
@@ -20895,7 +20937,7 @@
20895
20937
  "litellm_provider": "vercel_ai_gateway",
20896
20938
  "max_input_tokens": 128000,
20897
20939
  "max_output_tokens": 32768,
20898
- "max_tokens": 128000,
20940
+ "max_tokens": 32768,
20899
20941
  "mode": "chat",
20900
20942
  "output_cost_per_token": 1.5e-05
20901
20943
  },
@@ -20906,7 +20948,7 @@
20906
20948
  "litellm_provider": "vercel_ai_gateway",
20907
20949
  "max_input_tokens": 200000,
20908
20950
  "max_output_tokens": 100000,
20909
- "max_tokens": 200000,
20951
+ "max_tokens": 100000,
20910
20952
  "mode": "chat",
20911
20953
  "output_cost_per_token": 4.4e-06
20912
20954
  },
@@ -20915,7 +20957,7 @@
20915
20957
  "litellm_provider": "vercel_ai_gateway",
20916
20958
  "max_input_tokens": 128000,
20917
20959
  "max_output_tokens": 64000,
20918
- "max_tokens": 128000,
20960
+ "max_tokens": 64000,
20919
20961
  "mode": "chat",
20920
20962
  "output_cost_per_token": 5e-06
20921
20963
  },
@@ -20933,7 +20975,7 @@
20933
20975
  "litellm_provider": "vercel_ai_gateway",
20934
20976
  "max_input_tokens": 40960,
20935
20977
  "max_output_tokens": 16384,
20936
- "max_tokens": 40960,
20978
+ "max_tokens": 16384,
20937
20979
  "mode": "chat",
20938
20980
  "output_cost_per_token": 3e-07
20939
20981
  },
@@ -20942,7 +20984,7 @@
20942
20984
  "litellm_provider": "vercel_ai_gateway",
20943
20985
  "max_input_tokens": 128000,
20944
20986
  "max_output_tokens": 96000,
20945
- "max_tokens": 128000,
20987
+ "max_tokens": 96000,
20946
20988
  "mode": "chat",
20947
20989
  "output_cost_per_token": 1.1e-06
20948
20990
  },
@@ -20951,7 +20993,7 @@
20951
20993
  "litellm_provider": "vercel_ai_gateway",
20952
20994
  "max_input_tokens": 128000,
20953
20995
  "max_output_tokens": 4096,
20954
- "max_tokens": 128000,
20996
+ "max_tokens": 4096,
20955
20997
  "mode": "chat",
20956
20998
  "output_cost_per_token": 3e-05
20957
20999
  },
@@ -20960,7 +21002,7 @@
20960
21002
  "litellm_provider": "vercel_ai_gateway",
20961
21003
  "max_input_tokens": 32000,
20962
21004
  "max_output_tokens": 4000,
20963
- "max_tokens": 32000,
21005
+ "max_tokens": 4000,
20964
21006
  "mode": "chat",
20965
21007
  "output_cost_per_token": 6e-06
20966
21008
  },
@@ -20969,7 +21011,7 @@
20969
21011
  "litellm_provider": "vercel_ai_gateway",
20970
21012
  "max_input_tokens": 200000,
20971
21013
  "max_output_tokens": 8000,
20972
- "max_tokens": 200000,
21014
+ "max_tokens": 8000,
20973
21015
  "mode": "chat",
20974
21016
  "output_cost_per_token": 1.5e-05
20975
21017
  },
@@ -20978,7 +21020,7 @@
20978
21020
  "litellm_provider": "vercel_ai_gateway",
20979
21021
  "max_input_tokens": 128000,
20980
21022
  "max_output_tokens": 8192,
20981
- "max_tokens": 128000,
21023
+ "max_tokens": 8192,
20982
21024
  "mode": "chat",
20983
21025
  "output_cost_per_token": 7.2e-07
20984
21026
  },
@@ -21005,7 +21047,7 @@
21005
21047
  "litellm_provider": "vercel_ai_gateway",
21006
21048
  "max_input_tokens": 128000,
21007
21049
  "max_output_tokens": 4000,
21008
- "max_tokens": 128000,
21050
+ "max_tokens": 4000,
21009
21051
  "mode": "chat",
21010
21052
  "output_cost_per_token": 6e-06
21011
21053
  },
@@ -21016,7 +21058,7 @@
21016
21058
  "litellm_provider": "vercel_ai_gateway",
21017
21059
  "max_input_tokens": 200000,
21018
21060
  "max_output_tokens": 8192,
21019
- "max_tokens": 200000,
21061
+ "max_tokens": 8192,
21020
21062
  "mode": "chat",
21021
21063
  "output_cost_per_token": 1.5e-05
21022
21064
  },
@@ -21025,7 +21067,7 @@
21025
21067
  "litellm_provider": "vercel_ai_gateway",
21026
21068
  "max_input_tokens": 128000,
21027
21069
  "max_output_tokens": 8192,
21028
- "max_tokens": 128000,
21070
+ "max_tokens": 8192,
21029
21071
  "mode": "chat",
21030
21072
  "output_cost_per_token": 1.4e-07
21031
21073
  },
@@ -21034,7 +21076,7 @@
21034
21076
  "litellm_provider": "vercel_ai_gateway",
21035
21077
  "max_input_tokens": 128000,
21036
21078
  "max_output_tokens": 4096,
21037
- "max_tokens": 128000,
21079
+ "max_tokens": 4096,
21038
21080
  "mode": "chat",
21039
21081
  "output_cost_per_token": 6e-07
21040
21082
  },
@@ -21043,7 +21085,7 @@
21043
21085
  "litellm_provider": "vercel_ai_gateway",
21044
21086
  "max_input_tokens": 32768,
21045
21087
  "max_output_tokens": 16384,
21046
- "max_tokens": 32768,
21088
+ "max_tokens": 16384,
21047
21089
  "mode": "chat",
21048
21090
  "output_cost_per_token": 1.9e-06
21049
21091
  },
@@ -21052,7 +21094,7 @@
21052
21094
  "litellm_provider": "vercel_ai_gateway",
21053
21095
  "max_input_tokens": 65536,
21054
21096
  "max_output_tokens": 2048,
21055
- "max_tokens": 65536,
21097
+ "max_tokens": 2048,
21056
21098
  "mode": "chat",
21057
21099
  "output_cost_per_token": 1.2e-06
21058
21100
  },
@@ -21070,7 +21112,7 @@
21070
21112
  "litellm_provider": "vercel_ai_gateway",
21071
21113
  "max_input_tokens": 131000,
21072
21114
  "max_output_tokens": 131072,
21073
- "max_tokens": 131000,
21115
+ "max_tokens": 131072,
21074
21116
  "mode": "chat",
21075
21117
  "output_cost_per_token": 8e-08
21076
21118
  },
@@ -21081,7 +21123,7 @@
21081
21123
  "litellm_provider": "vercel_ai_gateway",
21082
21124
  "max_input_tokens": 200000,
21083
21125
  "max_output_tokens": 4096,
21084
- "max_tokens": 200000,
21126
+ "max_tokens": 4096,
21085
21127
  "mode": "chat",
21086
21128
  "output_cost_per_token": 7.5e-05
21087
21129
  },
@@ -21101,7 +21143,7 @@
21101
21143
  "litellm_provider": "vercel_ai_gateway",
21102
21144
  "max_input_tokens": 128000,
21103
21145
  "max_output_tokens": 16384,
21104
- "max_tokens": 128000,
21146
+ "max_tokens": 16384,
21105
21147
  "mode": "chat",
21106
21148
  "output_cost_per_token": 1e-05
21107
21149
  },
@@ -21112,7 +21154,7 @@
21112
21154
  "litellm_provider": "vercel_ai_gateway",
21113
21155
  "max_input_tokens": 200000,
21114
21156
  "max_output_tokens": 100000,
21115
- "max_tokens": 200000,
21157
+ "max_tokens": 100000,
21116
21158
  "mode": "chat",
21117
21159
  "output_cost_per_token": 4.4e-06
21118
21160
  },
@@ -21121,7 +21163,7 @@
21121
21163
  "litellm_provider": "vercel_ai_gateway",
21122
21164
  "max_input_tokens": 128000,
21123
21165
  "max_output_tokens": 4000,
21124
- "max_tokens": 128000,
21166
+ "max_tokens": 4000,
21125
21167
  "mode": "chat",
21126
21168
  "output_cost_per_token": 1e-07
21127
21169
  },
@@ -21132,7 +21174,7 @@
21132
21174
  "litellm_provider": "vercel_ai_gateway",
21133
21175
  "max_input_tokens": 200000,
21134
21176
  "max_output_tokens": 100000,
21135
- "max_tokens": 200000,
21177
+ "max_tokens": 100000,
21136
21178
  "mode": "chat",
21137
21179
  "output_cost_per_token": 8e-06
21138
21180
  },
@@ -21141,7 +21183,7 @@
21141
21183
  "litellm_provider": "vercel_ai_gateway",
21142
21184
  "max_input_tokens": 128000,
21143
21185
  "max_output_tokens": 32000,
21144
- "max_tokens": 128000,
21186
+ "max_tokens": 32000,
21145
21187
  "mode": "chat",
21146
21188
  "output_cost_per_token": 1.5e-05
21147
21189
  },
@@ -21159,7 +21201,7 @@
21159
21201
  "litellm_provider": "vercel_ai_gateway",
21160
21202
  "max_input_tokens": 300000,
21161
21203
  "max_output_tokens": 8192,
21162
- "max_tokens": 300000,
21204
+ "max_tokens": 8192,
21163
21205
  "mode": "chat",
21164
21206
  "output_cost_per_token": 3.2e-06
21165
21207
  },
@@ -21168,7 +21210,7 @@
21168
21210
  "litellm_provider": "vercel_ai_gateway",
21169
21211
  "max_input_tokens": 32768,
21170
21212
  "max_output_tokens": 16384,
21171
- "max_tokens": 32768,
21213
+ "max_tokens": 16384,
21172
21214
  "mode": "chat",
21173
21215
  "output_cost_per_token": 1.2e-06
21174
21216
  },
@@ -21177,7 +21219,7 @@
21177
21219
  "litellm_provider": "vercel_ai_gateway",
21178
21220
  "max_input_tokens": 16385,
21179
21221
  "max_output_tokens": 4096,
21180
- "max_tokens": 16385,
21222
+ "max_tokens": 4096,
21181
21223
  "mode": "chat",
21182
21224
  "output_cost_per_token": 1.5e-06
21183
21225
  },
@@ -21186,7 +21228,7 @@
21186
21228
  "litellm_provider": "vercel_ai_gateway",
21187
21229
  "max_input_tokens": 256000,
21188
21230
  "max_output_tokens": 4000,
21189
- "max_tokens": 256000,
21231
+ "max_tokens": 4000,
21190
21232
  "mode": "chat",
21191
21233
  "output_cost_per_token": 9e-07
21192
21234
  },
@@ -21195,7 +21237,7 @@
21195
21237
  "litellm_provider": "vercel_ai_gateway",
21196
21238
  "max_input_tokens": 128000,
21197
21239
  "max_output_tokens": 8192,
21198
- "max_tokens": 128000,
21240
+ "max_tokens": 8192,
21199
21241
  "mode": "chat",
21200
21242
  "output_cost_per_token": 1.6e-07
21201
21243
  },
@@ -21231,7 +21273,7 @@
21231
21273
  "litellm_provider": "vercel_ai_gateway",
21232
21274
  "max_input_tokens": 1048576,
21233
21275
  "max_output_tokens": 8192,
21234
- "max_tokens": 1048576,
21276
+ "max_tokens": 8192,
21235
21277
  "mode": "chat",
21236
21278
  "output_cost_per_token": 3e-07
21237
21279
  },
@@ -21240,7 +21282,7 @@
21240
21282
  "litellm_provider": "vercel_ai_gateway",
21241
21283
  "max_input_tokens": 128000,
21242
21284
  "max_output_tokens": 4000,
21243
- "max_tokens": 128000,
21285
+ "max_tokens": 4000,
21244
21286
  "mode": "chat",
21245
21287
  "output_cost_per_token": 4e-08
21246
21288
  },
@@ -21249,7 +21291,7 @@
21249
21291
  "litellm_provider": "vercel_ai_gateway",
21250
21292
  "max_input_tokens": 127000,
21251
21293
  "max_output_tokens": 8000,
21252
- "max_tokens": 127000,
21294
+ "max_tokens": 8000,
21253
21295
  "mode": "chat",
21254
21296
  "output_cost_per_token": 8e-06
21255
21297
  },
@@ -21269,7 +21311,7 @@
21269
21311
  "litellm_provider": "vercel_ai_gateway",
21270
21312
  "max_input_tokens": 200000,
21271
21313
  "max_output_tokens": 4096,
21272
- "max_tokens": 200000,
21314
+ "max_tokens": 4096,
21273
21315
  "mode": "chat",
21274
21316
  "output_cost_per_token": 1.25e-06
21275
21317
  },
@@ -21280,7 +21322,7 @@
21280
21322
  "litellm_provider": "vercel_ai_gateway",
21281
21323
  "max_input_tokens": 200000,
21282
21324
  "max_output_tokens": 100000,
21283
- "max_tokens": 200000,
21325
+ "max_tokens": 100000,
21284
21326
  "mode": "chat",
21285
21327
  "output_cost_per_token": 6e-05
21286
21328
  },
@@ -21289,7 +21331,7 @@
21289
21331
  "litellm_provider": "vercel_ai_gateway",
21290
21332
  "max_input_tokens": 128000,
21291
21333
  "max_output_tokens": 8192,
21292
- "max_tokens": 128000,
21334
+ "max_tokens": 8192,
21293
21335
  "mode": "chat",
21294
21336
  "output_cost_per_token": 2.19e-06
21295
21337
  },
@@ -21309,7 +21351,7 @@
21309
21351
  "litellm_provider": "vercel_ai_gateway",
21310
21352
  "max_input_tokens": 1047576,
21311
21353
  "max_output_tokens": 32768,
21312
- "max_tokens": 1047576,
21354
+ "max_tokens": 32768,
21313
21355
  "mode": "chat",
21314
21356
  "output_cost_per_token": 1.6e-06
21315
21357
  },
@@ -21320,7 +21362,7 @@
21320
21362
  "litellm_provider": "vercel_ai_gateway",
21321
21363
  "max_input_tokens": 128000,
21322
21364
  "max_output_tokens": 16384,
21323
- "max_tokens": 128000,
21365
+ "max_tokens": 16384,
21324
21366
  "mode": "chat",
21325
21367
  "output_cost_per_token": 6e-07
21326
21368
  },
@@ -21329,7 +21371,7 @@
21329
21371
  "litellm_provider": "vercel_ai_gateway",
21330
21372
  "max_input_tokens": 40960,
21331
21373
  "max_output_tokens": 16384,
21332
- "max_tokens": 40960,
21374
+ "max_tokens": 16384,
21333
21375
  "mode": "chat",
21334
21376
  "output_cost_per_token": 2.4e-07
21335
21377
  },
@@ -21340,7 +21382,7 @@
21340
21382
  "litellm_provider": "vercel_ai_gateway",
21341
21383
  "max_input_tokens": 200000,
21342
21384
  "max_output_tokens": 32000,
21343
- "max_tokens": 200000,
21385
+ "max_tokens": 32000,
21344
21386
  "mode": "chat",
21345
21387
  "output_cost_per_token": 7.5e-05
21346
21388
  },
@@ -21358,7 +21400,7 @@
21358
21400
  "litellm_provider": "vercel_ai_gateway",
21359
21401
  "max_input_tokens": 127000,
21360
21402
  "max_output_tokens": 8000,
21361
- "max_tokens": 127000,
21403
+ "max_tokens": 8000,
21362
21404
  "mode": "chat",
21363
21405
  "output_cost_per_token": 5e-06
21364
21406
  },
@@ -21369,7 +21411,7 @@
21369
21411
  "litellm_provider": "vercel_ai_gateway",
21370
21412
  "max_input_tokens": 200000,
21371
21413
  "max_output_tokens": 8192,
21372
- "max_tokens": 200000,
21414
+ "max_tokens": 8192,
21373
21415
  "mode": "chat",
21374
21416
  "output_cost_per_token": 4e-06
21375
21417
  },
@@ -21378,7 +21420,7 @@
21378
21420
  "litellm_provider": "vercel_ai_gateway",
21379
21421
  "max_input_tokens": 256000,
21380
21422
  "max_output_tokens": 8000,
21381
- "max_tokens": 256000,
21423
+ "max_tokens": 8000,
21382
21424
  "mode": "chat",
21383
21425
  "output_cost_per_token": 1e-05
21384
21426
  },
@@ -21396,7 +21438,7 @@
21396
21438
  "litellm_provider": "vercel_ai_gateway",
21397
21439
  "max_input_tokens": 128000,
21398
21440
  "max_output_tokens": 8192,
21399
- "max_tokens": 128000,
21441
+ "max_tokens": 8192,
21400
21442
  "mode": "chat",
21401
21443
  "output_cost_per_token": 1.5e-07
21402
21444
  },
@@ -21407,7 +21449,7 @@
21407
21449
  "litellm_provider": "vercel_ai_gateway",
21408
21450
  "max_input_tokens": 1047576,
21409
21451
  "max_output_tokens": 32768,
21410
- "max_tokens": 1047576,
21452
+ "max_tokens": 32768,
21411
21453
  "mode": "chat",
21412
21454
  "output_cost_per_token": 4e-07
21413
21455
  },
@@ -21418,7 +21460,7 @@
21418
21460
  "litellm_provider": "vercel_ai_gateway",
21419
21461
  "max_input_tokens": 200000,
21420
21462
  "max_output_tokens": 64000,
21421
- "max_tokens": 200000,
21463
+ "max_tokens": 64000,
21422
21464
  "mode": "chat",
21423
21465
  "output_cost_per_token": 1.5e-05
21424
21466
  },
@@ -21427,7 +21469,7 @@
21427
21469
  "litellm_provider": "vercel_ai_gateway",
21428
21470
  "max_input_tokens": 127000,
21429
21471
  "max_output_tokens": 8000,
21430
- "max_tokens": 127000,
21472
+ "max_tokens": 8000,
21431
21473
  "mode": "chat",
21432
21474
  "output_cost_per_token": 1e-06
21433
21475
  },
@@ -21436,7 +21478,7 @@
21436
21478
  "litellm_provider": "vercel_ai_gateway",
21437
21479
  "max_input_tokens": 131072,
21438
21480
  "max_output_tokens": 8192,
21439
- "max_tokens": 131072,
21481
+ "max_tokens": 8192,
21440
21482
  "mode": "chat",
21441
21483
  "output_cost_per_token": 6e-07
21442
21484
  },
@@ -21472,7 +21514,7 @@
21472
21514
  "litellm_provider": "vercel_ai_gateway",
21473
21515
  "max_input_tokens": 128000,
21474
21516
  "max_output_tokens": 8192,
21475
- "max_tokens": 128000,
21517
+ "max_tokens": 8192,
21476
21518
  "mode": "chat",
21477
21519
  "output_cost_per_token": 7.2e-07
21478
21520
  },
@@ -21481,7 +21523,7 @@
21481
21523
  "litellm_provider": "vercel_ai_gateway",
21482
21524
  "max_input_tokens": 128000,
21483
21525
  "max_output_tokens": 4096,
21484
- "max_tokens": 128000,
21526
+ "max_tokens": 4096,
21485
21527
  "mode": "chat",
21486
21528
  "output_cost_per_token": 1e-05
21487
21529
  },
@@ -21490,7 +21532,7 @@
21490
21532
  "litellm_provider": "vercel_ai_gateway",
21491
21533
  "max_input_tokens": 8192,
21492
21534
  "max_output_tokens": 4096,
21493
- "max_tokens": 8192,
21535
+ "max_tokens": 4096,
21494
21536
  "mode": "chat",
21495
21537
  "output_cost_per_token": 2e-06
21496
21538
  },
@@ -21510,7 +21552,7 @@
21510
21552
  "litellm_provider": "vercel_ai_gateway",
21511
21553
  "max_input_tokens": 200000,
21512
21554
  "max_output_tokens": 64000,
21513
- "max_tokens": 200000,
21555
+ "max_tokens": 64000,
21514
21556
  "mode": "chat",
21515
21557
  "output_cost_per_token": 1.5e-05
21516
21558
  },
@@ -21519,7 +21561,7 @@
21519
21561
  "litellm_provider": "vercel_ai_gateway",
21520
21562
  "max_input_tokens": 1048576,
21521
21563
  "max_output_tokens": 8192,
21522
- "max_tokens": 1048576,
21564
+ "max_tokens": 8192,
21523
21565
  "mode": "chat",
21524
21566
  "output_cost_per_token": 6e-07
21525
21567
  },
@@ -21528,7 +21570,7 @@
21528
21570
  "litellm_provider": "vercel_ai_gateway",
21529
21571
  "max_input_tokens": 128000,
21530
21572
  "max_output_tokens": 4000,
21531
- "max_tokens": 128000,
21573
+ "max_tokens": 4000,
21532
21574
  "mode": "chat",
21533
21575
  "output_cost_per_token": 1.5e-07
21534
21576
  },
@@ -21537,7 +21579,7 @@
21537
21579
  "litellm_provider": "vercel_ai_gateway",
21538
21580
  "max_input_tokens": 128000,
21539
21581
  "max_output_tokens": 64000,
21540
- "max_tokens": 128000,
21582
+ "max_tokens": 64000,
21541
21583
  "mode": "chat",
21542
21584
  "output_cost_per_token": 1.5e-06
21543
21585
  },
@@ -21546,7 +21588,7 @@
21546
21588
  "litellm_provider": "vercel_ai_gateway",
21547
21589
  "max_input_tokens": 131072,
21548
21590
  "max_output_tokens": 16384,
21549
- "max_tokens": 131072,
21591
+ "max_tokens": 16384,
21550
21592
  "mode": "chat",
21551
21593
  "output_cost_per_token": 2.2e-06
21552
21594
  },
@@ -21555,7 +21597,7 @@
21555
21597
  "litellm_provider": "vercel_ai_gateway",
21556
21598
  "max_input_tokens": 40960,
21557
21599
  "max_output_tokens": 16384,
21558
- "max_tokens": 40960,
21600
+ "max_tokens": 16384,
21559
21601
  "mode": "chat",
21560
21602
  "output_cost_per_token": 3e-07
21561
21603
  },
@@ -21566,7 +21608,7 @@
21566
21608
  "litellm_provider": "vercel_ai_gateway",
21567
21609
  "max_input_tokens": 1047576,
21568
21610
  "max_output_tokens": 32768,
21569
- "max_tokens": 1047576,
21611
+ "max_tokens": 32768,
21570
21612
  "mode": "chat",
21571
21613
  "output_cost_per_token": 8e-06
21572
21614
  },
@@ -21707,6 +21749,7 @@
21707
21749
  "supports_reasoning": true
21708
21750
  },
21709
21751
  "gemini/veo-3.0-generate-preview": {
21752
+ "deprecation_date": "2025-11-12",
21710
21753
  "litellm_provider": "gemini",
21711
21754
  "max_input_tokens": 1024,
21712
21755
  "max_tokens": 1024,
@@ -21721,6 +21764,7 @@
21721
21764
  ]
21722
21765
  },
21723
21766
  "gemini/veo-3.0-fast-generate-preview": {
21767
+ "deprecation_date": "2025-11-12",
21724
21768
  "litellm_provider": "gemini",
21725
21769
  "max_input_tokens": 1024,
21726
21770
  "max_tokens": 1024,
@@ -21749,6 +21793,7 @@
21749
21793
  ]
21750
21794
  },
21751
21795
  "vertex_ai/veo-3.0-generate-preview": {
21796
+ "deprecation_date": "2025-11-12",
21752
21797
  "litellm_provider": "vertex_ai-video-models",
21753
21798
  "max_input_tokens": 1024,
21754
21799
  "max_tokens": 1024,
@@ -21763,6 +21808,7 @@
21763
21808
  ]
21764
21809
  },
21765
21810
  "vertex_ai/veo-3.0-fast-generate-preview": {
21811
+ "deprecation_date": "2025-11-12",
21766
21812
  "litellm_provider": "vertex_ai-video-models",
21767
21813
  "max_input_tokens": 1024,
21768
21814
  "max_tokens": 1024,
@@ -21898,7 +21944,7 @@
21898
21944
  "litellm_provider": "dashscope",
21899
21945
  "max_input_tokens": 258048,
21900
21946
  "max_output_tokens": 65536,
21901
- "max_tokens": 262144,
21947
+ "max_tokens": 65536,
21902
21948
  "mode": "chat",
21903
21949
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
21904
21950
  "supports_function_calling": true,
@@ -21936,7 +21982,7 @@
21936
21982
  "litellm_provider": "dashscope",
21937
21983
  "max_input_tokens": 129024,
21938
21984
  "max_output_tokens": 16384,
21939
- "max_tokens": 131072,
21985
+ "max_tokens": 16384,
21940
21986
  "mode": "chat",
21941
21987
  "output_cost_per_token": 1.2e-06,
21942
21988
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -21948,7 +21994,7 @@
21948
21994
  "litellm_provider": "dashscope",
21949
21995
  "max_input_tokens": 997952,
21950
21996
  "max_output_tokens": 32768,
21951
- "max_tokens": 1000000,
21997
+ "max_tokens": 32768,
21952
21998
  "mode": "chat",
21953
21999
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
21954
22000
  "supports_function_calling": true,
@@ -21978,7 +22024,7 @@
21978
22024
  "litellm_provider": "dashscope",
21979
22025
  "max_input_tokens": 1000000,
21980
22026
  "max_output_tokens": 16384,
21981
- "max_tokens": 1000000,
22027
+ "max_tokens": 16384,
21982
22028
  "mode": "chat",
21983
22029
  "output_cost_per_token": 1.5e-06,
21984
22030
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -21990,7 +22036,7 @@
21990
22036
  "litellm_provider": "dashscope",
21991
22037
  "max_input_tokens": 997952,
21992
22038
  "max_output_tokens": 65536,
21993
- "max_tokens": 1000000,
22039
+ "max_tokens": 65536,
21994
22040
  "mode": "chat",
21995
22041
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
21996
22042
  "supports_function_calling": true,
@@ -22039,7 +22085,7 @@
22039
22085
  "litellm_provider": "dashscope",
22040
22086
  "max_input_tokens": 997952,
22041
22087
  "max_output_tokens": 65536,
22042
- "max_tokens": 1000000,
22088
+ "max_tokens": 65536,
22043
22089
  "mode": "chat",
22044
22090
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22045
22091
  "supports_function_calling": true,
@@ -22084,7 +22130,7 @@
22084
22130
  "litellm_provider": "dashscope",
22085
22131
  "max_input_tokens": 997952,
22086
22132
  "max_output_tokens": 65536,
22087
- "max_tokens": 1000000,
22133
+ "max_tokens": 65536,
22088
22134
  "mode": "chat",
22089
22135
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22090
22136
  "supports_function_calling": true,
@@ -22133,7 +22179,7 @@
22133
22179
  "litellm_provider": "dashscope",
22134
22180
  "max_input_tokens": 997952,
22135
22181
  "max_output_tokens": 65536,
22136
- "max_tokens": 1000000,
22182
+ "max_tokens": 65536,
22137
22183
  "mode": "chat",
22138
22184
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22139
22185
  "supports_function_calling": true,
@@ -22178,7 +22224,7 @@
22178
22224
  "litellm_provider": "dashscope",
22179
22225
  "max_input_tokens": 997952,
22180
22226
  "max_output_tokens": 32768,
22181
- "max_tokens": 1000000,
22227
+ "max_tokens": 32768,
22182
22228
  "mode": "chat",
22183
22229
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22184
22230
  "supports_function_calling": true,
@@ -22209,7 +22255,7 @@
22209
22255
  "litellm_provider": "dashscope",
22210
22256
  "max_input_tokens": 997952,
22211
22257
  "max_output_tokens": 32768,
22212
- "max_tokens": 1000000,
22258
+ "max_tokens": 32768,
22213
22259
  "mode": "chat",
22214
22260
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22215
22261
  "supports_function_calling": true,
@@ -22241,7 +22287,7 @@
22241
22287
  "litellm_provider": "dashscope",
22242
22288
  "max_input_tokens": 129024,
22243
22289
  "max_output_tokens": 16384,
22244
- "max_tokens": 131072,
22290
+ "max_tokens": 16384,
22245
22291
  "mode": "chat",
22246
22292
  "output_cost_per_reasoning_token": 4e-06,
22247
22293
  "output_cost_per_token": 1.2e-06,
@@ -22255,7 +22301,7 @@
22255
22301
  "litellm_provider": "dashscope",
22256
22302
  "max_input_tokens": 129024,
22257
22303
  "max_output_tokens": 16384,
22258
- "max_tokens": 131072,
22304
+ "max_tokens": 16384,
22259
22305
  "mode": "chat",
22260
22306
  "output_cost_per_reasoning_token": 4e-06,
22261
22307
  "output_cost_per_token": 1.2e-06,
@@ -22269,7 +22315,7 @@
22269
22315
  "litellm_provider": "dashscope",
22270
22316
  "max_input_tokens": 129024,
22271
22317
  "max_output_tokens": 8192,
22272
- "max_tokens": 131072,
22318
+ "max_tokens": 8192,
22273
22319
  "mode": "chat",
22274
22320
  "output_cost_per_token": 1.2e-06,
22275
22321
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -22281,7 +22327,7 @@
22281
22327
  "litellm_provider": "dashscope",
22282
22328
  "max_input_tokens": 997952,
22283
22329
  "max_output_tokens": 32768,
22284
- "max_tokens": 1000000,
22330
+ "max_tokens": 32768,
22285
22331
  "mode": "chat",
22286
22332
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
22287
22333
  "supports_function_calling": true,
@@ -22311,7 +22357,7 @@
22311
22357
  "litellm_provider": "dashscope",
22312
22358
  "max_input_tokens": 129024,
22313
22359
  "max_output_tokens": 16384,
22314
- "max_tokens": 131072,
22360
+ "max_tokens": 16384,
22315
22361
  "mode": "chat",
22316
22362
  "output_cost_per_reasoning_token": 5e-07,
22317
22363
  "output_cost_per_token": 2e-07,
@@ -22325,7 +22371,7 @@
22325
22371
  "litellm_provider": "dashscope",
22326
22372
  "max_input_tokens": 1000000,
22327
22373
  "max_output_tokens": 16384,
22328
- "max_tokens": 1000000,
22374
+ "max_tokens": 16384,
22329
22375
  "mode": "chat",
22330
22376
  "output_cost_per_reasoning_token": 5e-07,
22331
22377
  "output_cost_per_token": 2e-07,
@@ -22339,7 +22385,7 @@
22339
22385
  "litellm_provider": "dashscope",
22340
22386
  "max_input_tokens": 1000000,
22341
22387
  "max_output_tokens": 8192,
22342
- "max_tokens": 1000000,
22388
+ "max_tokens": 8192,
22343
22389
  "mode": "chat",
22344
22390
  "output_cost_per_token": 2e-07,
22345
22391
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -22352,7 +22398,7 @@
22352
22398
  "litellm_provider": "dashscope",
22353
22399
  "max_input_tokens": 98304,
22354
22400
  "max_output_tokens": 8192,
22355
- "max_tokens": 131072,
22401
+ "max_tokens": 8192,
22356
22402
  "mode": "chat",
22357
22403
  "output_cost_per_token": 2.4e-06,
22358
22404
  "source": "https://www.alibabacloud.com/help/en/model-studio/models",
@@ -22770,7 +22816,7 @@
22770
22816
  "litellm_provider": "openrouter",
22771
22817
  "max_input_tokens": 2000000,
22772
22818
  "max_output_tokens": 30000,
22773
- "max_tokens": 2000000,
22819
+ "max_tokens": 30000,
22774
22820
  "mode": "chat",
22775
22821
  "output_cost_per_token": 0,
22776
22822
  "source": "https://openrouter.ai/x-ai/grok-4-fast:free",
@@ -22871,7 +22917,7 @@
22871
22917
  "litellm_provider": "deepseek",
22872
22918
  "max_input_tokens": 131072,
22873
22919
  "max_output_tokens": 65536,
22874
- "max_tokens": 131072,
22920
+ "max_tokens": 65536,
22875
22921
  "mode": "chat",
22876
22922
  "output_cost_per_token": 1.7e-06,
22877
22923
  "source": "https://api-docs.deepseek.com/quick_start/pricing",
@@ -22892,7 +22938,7 @@
22892
22938
  "litellm_provider": "bedrock_converse",
22893
22939
  "max_input_tokens": 163840,
22894
22940
  "max_output_tokens": 81920,
22895
- "max_tokens": 163840,
22941
+ "max_tokens": 81920,
22896
22942
  "mode": "chat",
22897
22943
  "output_cost_per_token": 1.68e-06,
22898
22944
  "supports_function_calling": true,
@@ -22934,7 +22980,7 @@
22934
22980
  "litellm_provider": "bedrock_converse",
22935
22981
  "max_input_tokens": 262000,
22936
22982
  "max_output_tokens": 65536,
22937
- "max_tokens": 262144,
22983
+ "max_tokens": 65536,
22938
22984
  "mode": "chat",
22939
22985
  "output_cost_per_token": 1.8e-06,
22940
22986
  "supports_function_calling": true,
@@ -22946,7 +22992,7 @@
22946
22992
  "litellm_provider": "bedrock_converse",
22947
22993
  "max_input_tokens": 262144,
22948
22994
  "max_output_tokens": 131072,
22949
- "max_tokens": 262144,
22995
+ "max_tokens": 131072,
22950
22996
  "mode": "chat",
22951
22997
  "output_cost_per_token": 8.8e-07,
22952
22998
  "supports_function_calling": true,
@@ -22958,7 +23004,7 @@
22958
23004
  "litellm_provider": "bedrock_converse",
22959
23005
  "max_input_tokens": 262144,
22960
23006
  "max_output_tokens": 131072,
22961
- "max_tokens": 262144,
23007
+ "max_tokens": 131072,
22962
23008
  "mode": "chat",
22963
23009
  "output_cost_per_token": 6e-07,
22964
23010
  "supports_function_calling": true,
@@ -22970,7 +23016,7 @@
22970
23016
  "litellm_provider": "bedrock_converse",
22971
23017
  "max_input_tokens": 131072,
22972
23018
  "max_output_tokens": 16384,
22973
- "max_tokens": 131072,
23019
+ "max_tokens": 16384,
22974
23020
  "mode": "chat",
22975
23021
  "output_cost_per_token": 6e-07,
22976
23022
  "supports_function_calling": true,
@@ -23008,7 +23054,7 @@
23008
23054
  "litellm_provider": "vertex_ai-deepseek_models",
23009
23055
  "max_input_tokens": 163840,
23010
23056
  "max_output_tokens": 32768,
23011
- "max_tokens": 163840,
23057
+ "max_tokens": 32768,
23012
23058
  "mode": "chat",
23013
23059
  "output_cost_per_token": 5.4e-06,
23014
23060
  "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models",
@@ -23639,7 +23685,7 @@
23639
23685
  "lemonade/Qwen3-Coder-30B-A3B-Instruct-GGUF": {
23640
23686
  "input_cost_per_token": 0,
23641
23687
  "litellm_provider": "lemonade",
23642
- "max_tokens": 262144,
23688
+ "max_tokens": 32768,
23643
23689
  "max_input_tokens": 262144,
23644
23690
  "max_output_tokens": 32768,
23645
23691
  "mode": "chat",
@@ -23655,7 +23701,7 @@
23655
23701
  "litellm_provider": "groq",
23656
23702
  "max_input_tokens": 262144,
23657
23703
  "max_output_tokens": 16384,
23658
- "max_tokens": 278528,
23704
+ "max_tokens": 16384,
23659
23705
  "mode": "chat",
23660
23706
  "supports_function_calling": true,
23661
23707
  "supports_response_schema": true,
@@ -24451,7 +24497,7 @@
24451
24497
  "litellm_provider": "oci",
24452
24498
  "max_input_tokens": 128000,
24453
24499
  "max_output_tokens": 4000,
24454
- "max_tokens": 128000,
24500
+ "max_tokens": 4000,
24455
24501
  "mode": "chat",
24456
24502
  "output_cost_per_token": 1.56e-06,
24457
24503
  "source": "https://www.oracle.com/cloud/ai/generative-ai/pricing/",
@@ -24463,7 +24509,7 @@
24463
24509
  "litellm_provider": "oci",
24464
24510
  "max_input_tokens": 256000,
24465
24511
  "max_output_tokens": 4000,
24466
- "max_tokens": 256000,
24512
+ "max_tokens": 4000,
24467
24513
  "mode": "chat",
24468
24514
  "output_cost_per_token": 1.56e-06,
24469
24515
  "source": "https://www.oracle.com/cloud/ai/generative-ai/pricing/",
@@ -24475,7 +24521,7 @@
24475
24521
  "litellm_provider": "oci",
24476
24522
  "max_input_tokens": 128000,
24477
24523
  "max_output_tokens": 4000,
24478
- "max_tokens": 128000,
24524
+ "max_tokens": 4000,
24479
24525
  "mode": "chat",
24480
24526
  "output_cost_per_token": 1.56e-06,
24481
24527
  "source": "https://www.oracle.com/cloud/ai/generative-ai/pricing/",
@@ -24790,7 +24836,7 @@
24790
24836
  "lemonade/gpt-oss-20b-mxfp4-GGUF": {
24791
24837
  "input_cost_per_token": 0,
24792
24838
  "litellm_provider": "lemonade",
24793
- "max_tokens": 131072,
24839
+ "max_tokens": 32768,
24794
24840
  "max_input_tokens": 131072,
24795
24841
  "max_output_tokens": 32768,
24796
24842
  "mode": "chat",
@@ -24802,7 +24848,7 @@
24802
24848
  "lemonade/gpt-oss-120b-mxfp-GGUF": {
24803
24849
  "input_cost_per_token": 0,
24804
24850
  "litellm_provider": "lemonade",
24805
- "max_tokens": 131072,
24851
+ "max_tokens": 32768,
24806
24852
  "max_input_tokens": 131072,
24807
24853
  "max_output_tokens": 32768,
24808
24854
  "mode": "chat",
@@ -24814,7 +24860,7 @@
24814
24860
  "lemonade/Gemma-3-4b-it-GGUF": {
24815
24861
  "input_cost_per_token": 0,
24816
24862
  "litellm_provider": "lemonade",
24817
- "max_tokens": 128000,
24863
+ "max_tokens": 8192,
24818
24864
  "max_input_tokens": 128000,
24819
24865
  "max_output_tokens": 8192,
24820
24866
  "mode": "chat",
@@ -24826,7 +24872,7 @@
24826
24872
  "lemonade/Qwen3-4B-Instruct-2507-GGUF": {
24827
24873
  "input_cost_per_token": 0,
24828
24874
  "litellm_provider": "lemonade",
24829
- "max_tokens": 262144,
24875
+ "max_tokens": 32768,
24830
24876
  "max_input_tokens": 262144,
24831
24877
  "max_output_tokens": 32768,
24832
24878
  "mode": "chat",
@@ -25559,12 +25605,16 @@
25559
25605
  "tpm": 8000000
25560
25606
  },
25561
25607
  "azure/gpt-image-1-mini": {
25562
- "input_cost_per_pixel": 8.0566406e-09,
25608
+ "cache_read_input_image_token_cost": 2.5e-07,
25609
+ "cache_read_input_token_cost": 2e-07,
25610
+ "input_cost_per_image_token": 2.5e-06,
25611
+ "input_cost_per_token": 2e-06,
25563
25612
  "litellm_provider": "azure",
25564
25613
  "mode": "image_generation",
25565
- "output_cost_per_pixel": 0.0,
25614
+ "output_cost_per_image_token": 8e-06,
25566
25615
  "supported_endpoints": [
25567
- "/v1/images/generations"
25616
+ "/v1/images/generations",
25617
+ "/v1/images/edits"
25568
25618
  ]
25569
25619
  },
25570
25620
  "azure/low/1024-x-1024/gpt-image-1-mini": {
@@ -25740,7 +25790,7 @@
25740
25790
  "litellm_provider": "azure",
25741
25791
  "max_input_tokens": 272000,
25742
25792
  "max_output_tokens": 128000,
25743
- "max_tokens": 400000,
25793
+ "max_tokens": 128000,
25744
25794
  "mode": "responses",
25745
25795
  "output_cost_per_token": 0.00012,
25746
25796
  "source": "https://learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure?pivots=azure-openai&tabs=global-standard-aoai%2Cstandard-chat-completions%2Cglobal-standard#gpt-5",
@@ -25877,7 +25927,7 @@
25877
25927
  "litellm_provider": "openrouter",
25878
25928
  "max_input_tokens": 163840,
25879
25929
  "max_output_tokens": 163840,
25880
- "max_tokens": 8192,
25930
+ "max_tokens": 163840,
25881
25931
  "mode": "chat",
25882
25932
  "output_cost_per_token": 4e-07,
25883
25933
  "supports_assistant_prefill": true,
@@ -25891,7 +25941,7 @@
25891
25941
  "litellm_provider": "openrouter",
25892
25942
  "max_input_tokens": 204800,
25893
25943
  "max_output_tokens": 204800,
25894
- "max_tokens": 32768,
25944
+ "max_tokens": 204800,
25895
25945
  "mode": "chat",
25896
25946
  "output_cost_per_token": 1.02e-06,
25897
25947
  "supports_function_calling": true,
@@ -25904,7 +25954,7 @@
25904
25954
  "litellm_provider": "openrouter",
25905
25955
  "max_input_tokens": 202800,
25906
25956
  "max_output_tokens": 131000,
25907
- "max_tokens": 202800,
25957
+ "max_tokens": 131000,
25908
25958
  "mode": "chat",
25909
25959
  "output_cost_per_token": 1.75e-06,
25910
25960
  "source": "https://openrouter.ai/z-ai/glm-4.6",
@@ -25917,7 +25967,7 @@
25917
25967
  "litellm_provider": "openrouter",
25918
25968
  "max_input_tokens": 202800,
25919
25969
  "max_output_tokens": 131000,
25920
- "max_tokens": 202800,
25970
+ "max_tokens": 131000,
25921
25971
  "mode": "chat",
25922
25972
  "output_cost_per_token": 1.9e-06,
25923
25973
  "source": "https://openrouter.ai/z-ai/glm-4.6:exacto",
@@ -27690,7 +27740,7 @@
27690
27740
  "litellm_provider": "vertex_ai-language-models",
27691
27741
  "max_input_tokens": 65536,
27692
27742
  "max_output_tokens": 32768,
27693
- "max_tokens": 65536,
27743
+ "max_tokens": 32768,
27694
27744
  "mode": "image_generation",
27695
27745
  "output_cost_per_image": 0.134,
27696
27746
  "output_cost_per_image_token": 0.00012,
@@ -27724,7 +27774,7 @@
27724
27774
  "litellm_provider": "gemini",
27725
27775
  "max_input_tokens": 65536,
27726
27776
  "max_output_tokens": 32768,
27727
- "max_tokens": 65536,
27777
+ "max_tokens": 32768,
27728
27778
  "mode": "image_generation",
27729
27779
  "output_cost_per_image": 0.134,
27730
27780
  "output_cost_per_image_token": 0.00012,
@@ -27821,7 +27871,7 @@
27821
27871
  "litellm_provider": "vertex_ai-language-models",
27822
27872
  "max_input_tokens": 65536,
27823
27873
  "max_output_tokens": 32768,
27824
- "max_tokens": 65536,
27874
+ "max_tokens": 32768,
27825
27875
  "mode": "image_generation",
27826
27876
  "output_cost_per_image": 0.134,
27827
27877
  "output_cost_per_image_token": 0.00012,
@@ -28265,7 +28315,7 @@
28265
28315
  "litellm_provider": "publicai",
28266
28316
  "max_input_tokens": 8192,
28267
28317
  "max_output_tokens": 4096,
28268
- "max_tokens": 8192,
28318
+ "max_tokens": 4096,
28269
28319
  "mode": "chat",
28270
28320
  "output_cost_per_token": 0.0,
28271
28321
  "source": "https://platform.publicai.co/docs",
@@ -28277,7 +28327,7 @@
28277
28327
  "litellm_provider": "publicai",
28278
28328
  "max_input_tokens": 8192,
28279
28329
  "max_output_tokens": 4096,
28280
- "max_tokens": 8192,
28330
+ "max_tokens": 4096,
28281
28331
  "mode": "chat",
28282
28332
  "output_cost_per_token": 0.0,
28283
28333
  "source": "https://platform.publicai.co/docs",
@@ -28289,7 +28339,7 @@
28289
28339
  "litellm_provider": "publicai",
28290
28340
  "max_input_tokens": 8192,
28291
28341
  "max_output_tokens": 4096,
28292
- "max_tokens": 8192,
28342
+ "max_tokens": 4096,
28293
28343
  "mode": "chat",
28294
28344
  "output_cost_per_token": 0.0,
28295
28345
  "source": "https://platform.publicai.co/docs",
@@ -28301,7 +28351,7 @@
28301
28351
  "litellm_provider": "publicai",
28302
28352
  "max_input_tokens": 16384,
28303
28353
  "max_output_tokens": 4096,
28304
- "max_tokens": 16384,
28354
+ "max_tokens": 4096,
28305
28355
  "mode": "chat",
28306
28356
  "output_cost_per_token": 0.0,
28307
28357
  "source": "https://platform.publicai.co/docs",
@@ -28313,7 +28363,7 @@
28313
28363
  "litellm_provider": "publicai",
28314
28364
  "max_input_tokens": 8192,
28315
28365
  "max_output_tokens": 4096,
28316
- "max_tokens": 8192,
28366
+ "max_tokens": 4096,
28317
28367
  "mode": "chat",
28318
28368
  "output_cost_per_token": 0.0,
28319
28369
  "source": "https://platform.publicai.co/docs",
@@ -28325,7 +28375,7 @@
28325
28375
  "litellm_provider": "publicai",
28326
28376
  "max_input_tokens": 32768,
28327
28377
  "max_output_tokens": 4096,
28328
- "max_tokens": 32768,
28378
+ "max_tokens": 4096,
28329
28379
  "mode": "chat",
28330
28380
  "output_cost_per_token": 0.0,
28331
28381
  "source": "https://platform.publicai.co/docs",
@@ -28337,7 +28387,7 @@
28337
28387
  "litellm_provider": "publicai",
28338
28388
  "max_input_tokens": 32768,
28339
28389
  "max_output_tokens": 4096,
28340
- "max_tokens": 32768,
28390
+ "max_tokens": 4096,
28341
28391
  "mode": "chat",
28342
28392
  "output_cost_per_token": 0.0,
28343
28393
  "source": "https://platform.publicai.co/docs",
@@ -28349,7 +28399,7 @@
28349
28399
  "litellm_provider": "publicai",
28350
28400
  "max_input_tokens": 32768,
28351
28401
  "max_output_tokens": 4096,
28352
- "max_tokens": 32768,
28402
+ "max_tokens": 4096,
28353
28403
  "mode": "chat",
28354
28404
  "output_cost_per_token": 0.0,
28355
28405
  "source": "https://platform.publicai.co/docs",
@@ -28362,7 +28412,7 @@
28362
28412
  "litellm_provider": "publicai",
28363
28413
  "max_input_tokens": 32768,
28364
28414
  "max_output_tokens": 4096,
28365
- "max_tokens": 32768,
28415
+ "max_tokens": 4096,
28366
28416
  "mode": "chat",
28367
28417
  "output_cost_per_token": 0.0,
28368
28418
  "source": "https://platform.publicai.co/docs",
@@ -28375,7 +28425,7 @@
28375
28425
  "litellm_provider": "fireworks_ai",
28376
28426
  "max_input_tokens": 262144,
28377
28427
  "max_output_tokens": 32768,
28378
- "max_tokens": 262144,
28428
+ "max_tokens": 32768,
28379
28429
  "mode": "chat",
28380
28430
  "output_cost_per_token": 2.5e-06,
28381
28431
  "source": "https://app.fireworks.ai/models/fireworks/kimi-k2-instruct-0905",
@@ -28416,7 +28466,7 @@
28416
28466
  "litellm_provider": "databricks",
28417
28467
  "max_input_tokens": 200000,
28418
28468
  "max_output_tokens": 64000,
28419
- "max_tokens": 200000,
28469
+ "max_tokens": 64000,
28420
28470
  "metadata": {
28421
28471
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28422
28472
  },
@@ -28435,7 +28485,7 @@
28435
28485
  "litellm_provider": "databricks",
28436
28486
  "max_input_tokens": 200000,
28437
28487
  "max_output_tokens": 32000,
28438
- "max_tokens": 200000,
28488
+ "max_tokens": 32000,
28439
28489
  "metadata": {
28440
28490
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28441
28491
  },
@@ -28454,7 +28504,7 @@
28454
28504
  "litellm_provider": "databricks",
28455
28505
  "max_input_tokens": 200000,
28456
28506
  "max_output_tokens": 32000,
28457
- "max_tokens": 200000,
28507
+ "max_tokens": 32000,
28458
28508
  "metadata": {
28459
28509
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28460
28510
  },
@@ -28473,7 +28523,7 @@
28473
28523
  "litellm_provider": "databricks",
28474
28524
  "max_input_tokens": 200000,
28475
28525
  "max_output_tokens": 64000,
28476
- "max_tokens": 200000,
28526
+ "max_tokens": 64000,
28477
28527
  "metadata": {
28478
28528
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28479
28529
  },
@@ -28492,7 +28542,7 @@
28492
28542
  "litellm_provider": "databricks",
28493
28543
  "max_input_tokens": 200000,
28494
28544
  "max_output_tokens": 64000,
28495
- "max_tokens": 200000,
28545
+ "max_tokens": 64000,
28496
28546
  "metadata": {
28497
28547
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28498
28548
  },
@@ -28511,7 +28561,7 @@
28511
28561
  "litellm_provider": "databricks",
28512
28562
  "max_input_tokens": 200000,
28513
28563
  "max_output_tokens": 64000,
28514
- "max_tokens": 200000,
28564
+ "max_tokens": 64000,
28515
28565
  "metadata": {
28516
28566
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28517
28567
  },
@@ -28530,7 +28580,7 @@
28530
28580
  "litellm_provider": "databricks",
28531
28581
  "max_input_tokens": 200000,
28532
28582
  "max_output_tokens": 64000,
28533
- "max_tokens": 200000,
28583
+ "max_tokens": 64000,
28534
28584
  "metadata": {
28535
28585
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28536
28586
  },
@@ -28549,7 +28599,7 @@
28549
28599
  "litellm_provider": "databricks",
28550
28600
  "max_input_tokens": 1048576,
28551
28601
  "max_output_tokens": 65535,
28552
- "max_tokens": 1048576,
28602
+ "max_tokens": 65535,
28553
28603
  "metadata": {
28554
28604
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28555
28605
  },
@@ -28566,7 +28616,7 @@
28566
28616
  "litellm_provider": "databricks",
28567
28617
  "max_input_tokens": 1048576,
28568
28618
  "max_output_tokens": 65536,
28569
- "max_tokens": 1048576,
28619
+ "max_tokens": 65536,
28570
28620
  "metadata": {
28571
28621
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28572
28622
  },
@@ -28583,7 +28633,7 @@
28583
28633
  "litellm_provider": "databricks",
28584
28634
  "max_input_tokens": 128000,
28585
28635
  "max_output_tokens": 32000,
28586
- "max_tokens": 128000,
28636
+ "max_tokens": 32000,
28587
28637
  "metadata": {
28588
28638
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28589
28639
  },
@@ -28598,7 +28648,7 @@
28598
28648
  "litellm_provider": "databricks",
28599
28649
  "max_input_tokens": 400000,
28600
28650
  "max_output_tokens": 128000,
28601
- "max_tokens": 400000,
28651
+ "max_tokens": 128000,
28602
28652
  "metadata": {
28603
28653
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28604
28654
  },
@@ -28613,7 +28663,7 @@
28613
28663
  "litellm_provider": "databricks",
28614
28664
  "max_input_tokens": 400000,
28615
28665
  "max_output_tokens": 128000,
28616
- "max_tokens": 400000,
28666
+ "max_tokens": 128000,
28617
28667
  "metadata": {
28618
28668
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28619
28669
  },
@@ -28628,7 +28678,7 @@
28628
28678
  "litellm_provider": "databricks",
28629
28679
  "max_input_tokens": 400000,
28630
28680
  "max_output_tokens": 128000,
28631
- "max_tokens": 400000,
28681
+ "max_tokens": 128000,
28632
28682
  "metadata": {
28633
28683
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28634
28684
  },
@@ -28643,7 +28693,7 @@
28643
28693
  "litellm_provider": "databricks",
28644
28694
  "max_input_tokens": 400000,
28645
28695
  "max_output_tokens": 128000,
28646
- "max_tokens": 400000,
28696
+ "max_tokens": 128000,
28647
28697
  "metadata": {
28648
28698
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28649
28699
  },
@@ -28688,7 +28738,7 @@
28688
28738
  "litellm_provider": "databricks",
28689
28739
  "max_input_tokens": 200000,
28690
28740
  "max_output_tokens": 128000,
28691
- "max_tokens": 200000,
28741
+ "max_tokens": 128000,
28692
28742
  "metadata": {
28693
28743
  "notes": "Input/output cost per token is dbu cost * $0.070. Number provided for reference, '*_dbu_cost_per_token' used in actual calculation."
28694
28744
  },
@@ -28698,13 +28748,13 @@
28698
28748
  "source": "https://www.databricks.com/product/pricing/foundation-model-serving"
28699
28749
  },
28700
28750
  "fireworks_ai/accounts/fireworks/models/deepseek-v3p2": {
28701
- "input_cost_per_token": 1.2e-06,
28751
+ "input_cost_per_token": 5.6e-07,
28702
28752
  "litellm_provider": "fireworks_ai",
28703
28753
  "max_input_tokens": 163840,
28704
28754
  "max_output_tokens": 163840,
28705
28755
  "max_tokens": 163840,
28706
28756
  "mode": "chat",
28707
- "output_cost_per_token": 1.2e-06,
28757
+ "output_cost_per_token": 1.68e-06,
28708
28758
  "source": "https://fireworks.ai/models/fireworks/deepseek-v3p2",
28709
28759
  "supports_function_calling": true,
28710
28760
  "supports_reasoning": true,
@@ -28883,7 +28933,7 @@
28883
28933
  "litellm_provider": "deepseek",
28884
28934
  "max_input_tokens": 163840,
28885
28935
  "max_output_tokens": 163840,
28886
- "max_tokens": 8192,
28936
+ "max_tokens": 163840,
28887
28937
  "mode": "chat",
28888
28938
  "output_cost_per_token": 4e-07,
28889
28939
  "supports_assistant_prefill": true,
@@ -28970,7 +29020,7 @@
28970
29020
  "litellm_provider": "openrouter",
28971
29021
  "max_input_tokens": 163840,
28972
29022
  "max_output_tokens": 163840,
28973
- "max_tokens": 8192,
29023
+ "max_tokens": 163840,
28974
29024
  "mode": "chat",
28975
29025
  "output_cost_per_token": 4e-07,
28976
29026
  "supports_assistant_prefill": true,
@@ -31559,7 +31609,7 @@
31559
31609
  "litellm_provider": "vertex_ai-deepseek_models",
31560
31610
  "max_input_tokens": 163840,
31561
31611
  "max_output_tokens": 32768,
31562
- "max_tokens": 163840,
31612
+ "max_tokens": 32768,
31563
31613
  "mode": "chat",
31564
31614
  "output_cost_per_token": 1.68e-06,
31565
31615
  "output_cost_per_token_batches": 8.4e-07,
@@ -32165,7 +32215,7 @@
32165
32215
  "litellm_provider": "openrouter",
32166
32216
  "max_input_tokens": 262144,
32167
32217
  "max_output_tokens": 65536,
32168
- "max_tokens": 262144,
32218
+ "max_tokens": 65536,
32169
32219
  "mode": "chat",
32170
32220
  "output_cost_per_token": 6e-07,
32171
32221
  "supports_function_calling": true,
@@ -32236,7 +32286,7 @@
32236
32286
  "litellm_provider": "openrouter",
32237
32287
  "max_input_tokens": 400000,
32238
32288
  "max_output_tokens": 128000,
32239
- "max_tokens": 400000,
32289
+ "max_tokens": 128000,
32240
32290
  "mode": "chat",
32241
32291
  "output_cost_per_token": 1.4e-05,
32242
32292
  "supports_function_calling": true,
@@ -32252,7 +32302,7 @@
32252
32302
  "litellm_provider": "openrouter",
32253
32303
  "max_input_tokens": 128000,
32254
32304
  "max_output_tokens": 16384,
32255
- "max_tokens": 128000,
32305
+ "max_tokens": 16384,
32256
32306
  "mode": "chat",
32257
32307
  "output_cost_per_token": 1.4e-05,
32258
32308
  "supports_function_calling": true,
@@ -32266,7 +32316,7 @@
32266
32316
  "litellm_provider": "openrouter",
32267
32317
  "max_input_tokens": 400000,
32268
32318
  "max_output_tokens": 128000,
32269
- "max_tokens": 400000,
32319
+ "max_tokens": 128000,
32270
32320
  "mode": "chat",
32271
32321
  "output_cost_per_token": 0.000168,
32272
32322
  "supports_function_calling": true,
@@ -32280,7 +32330,7 @@
32280
32330
  "litellm_provider": "azure_ai",
32281
32331
  "max_input_tokens": 163840,
32282
32332
  "max_output_tokens": 163840,
32283
- "max_tokens": 8192,
32333
+ "max_tokens": 163840,
32284
32334
  "mode": "chat",
32285
32335
  "output_cost_per_token": 1.68e-06,
32286
32336
  "supports_assistant_prefill": true,
@@ -32294,7 +32344,7 @@
32294
32344
  "litellm_provider": "azure_ai",
32295
32345
  "max_input_tokens": 163840,
32296
32346
  "max_output_tokens": 163840,
32297
- "max_tokens": 8192,
32347
+ "max_tokens": 163840,
32298
32348
  "mode": "chat",
32299
32349
  "output_cost_per_token": 1.68e-06,
32300
32350
  "supports_assistant_prefill": true,
@@ -33240,5 +33290,2084 @@
33240
33290
  "supports_system_messages": true,
33241
33291
  "supports_tool_choice": true,
33242
33292
  "supports_vision": true
33293
+ },
33294
+ "aiml/google/imagen-4.0-ultra-generate-001": {
33295
+ "litellm_provider": "aiml",
33296
+ "metadata": {
33297
+ "notes": "Imagen 4.0 Ultra Generate API - Photorealistic image generation with precise text rendering"
33298
+ },
33299
+ "mode": "image_generation",
33300
+ "output_cost_per_image": 0.063,
33301
+ "source": "https://docs.aimlapi.com/api-references/image-models/google/imagen-4-ultra-generate",
33302
+ "supported_endpoints": [
33303
+ "/v1/images/generations"
33304
+ ]
33305
+ },
33306
+ "aiml/google/nano-banana-pro": {
33307
+ "litellm_provider": "aiml",
33308
+ "metadata": {
33309
+ "notes": "Gemini 3 Pro Image (Nano Banana Pro) - Advanced text-to-image generation with reasoning and 4K resolution support"
33310
+ },
33311
+ "mode": "image_generation",
33312
+ "output_cost_per_image": 0.1575,
33313
+ "source": "https://docs.aimlapi.com/api-references/image-models/google/gemini-3-pro-image-preview",
33314
+ "supported_endpoints": [
33315
+ "/v1/images/generations"
33316
+ ]
33317
+ },
33318
+ "low/1024-x-1024/gpt-image-1.5": {
33319
+ "input_cost_per_image": 0.009,
33320
+ "litellm_provider": "openai",
33321
+ "mode": "image_generation",
33322
+ "supported_endpoints": [
33323
+ "/v1/images/generations",
33324
+ "/v1/images/edits"
33325
+ ],
33326
+ "supports_vision": true,
33327
+ "supports_pdf_input": true
33328
+ },
33329
+ "low/1024-x-1536/gpt-image-1.5": {
33330
+ "input_cost_per_image": 0.013,
33331
+ "litellm_provider": "openai",
33332
+ "mode": "image_generation",
33333
+ "supported_endpoints": [
33334
+ "/v1/images/generations",
33335
+ "/v1/images/edits"
33336
+ ],
33337
+ "supports_vision": true,
33338
+ "supports_pdf_input": true
33339
+ },
33340
+ "low/1536-x-1024/gpt-image-1.5": {
33341
+ "input_cost_per_image": 0.013,
33342
+ "litellm_provider": "openai",
33343
+ "mode": "image_generation",
33344
+ "supported_endpoints": [
33345
+ "/v1/images/generations",
33346
+ "/v1/images/edits"
33347
+ ],
33348
+ "supports_vision": true,
33349
+ "supports_pdf_input": true
33350
+ },
33351
+ "medium/1024-x-1024/gpt-image-1.5": {
33352
+ "input_cost_per_image": 0.034,
33353
+ "litellm_provider": "openai",
33354
+ "mode": "image_generation",
33355
+ "supported_endpoints": [
33356
+ "/v1/images/generations",
33357
+ "/v1/images/edits"
33358
+ ],
33359
+ "supports_vision": true,
33360
+ "supports_pdf_input": true
33361
+ },
33362
+ "medium/1024-x-1536/gpt-image-1.5": {
33363
+ "input_cost_per_image": 0.05,
33364
+ "litellm_provider": "openai",
33365
+ "mode": "image_generation",
33366
+ "supported_endpoints": [
33367
+ "/v1/images/generations",
33368
+ "/v1/images/edits"
33369
+ ],
33370
+ "supports_vision": true,
33371
+ "supports_pdf_input": true
33372
+ },
33373
+ "medium/1536-x-1024/gpt-image-1.5": {
33374
+ "input_cost_per_image": 0.05,
33375
+ "litellm_provider": "openai",
33376
+ "mode": "image_generation",
33377
+ "supported_endpoints": [
33378
+ "/v1/images/generations",
33379
+ "/v1/images/edits"
33380
+ ],
33381
+ "supports_vision": true,
33382
+ "supports_pdf_input": true
33383
+ },
33384
+ "high/1024-x-1024/gpt-image-1.5": {
33385
+ "input_cost_per_image": 0.133,
33386
+ "litellm_provider": "openai",
33387
+ "mode": "image_generation",
33388
+ "supported_endpoints": [
33389
+ "/v1/images/generations",
33390
+ "/v1/images/edits"
33391
+ ],
33392
+ "supports_vision": true,
33393
+ "supports_pdf_input": true
33394
+ },
33395
+ "high/1024-x-1536/gpt-image-1.5": {
33396
+ "input_cost_per_image": 0.2,
33397
+ "litellm_provider": "openai",
33398
+ "mode": "image_generation",
33399
+ "supported_endpoints": [
33400
+ "/v1/images/generations",
33401
+ "/v1/images/edits"
33402
+ ],
33403
+ "supports_vision": true,
33404
+ "supports_pdf_input": true
33405
+ },
33406
+ "high/1536-x-1024/gpt-image-1.5": {
33407
+ "input_cost_per_image": 0.2,
33408
+ "litellm_provider": "openai",
33409
+ "mode": "image_generation",
33410
+ "supported_endpoints": [
33411
+ "/v1/images/generations",
33412
+ "/v1/images/edits"
33413
+ ],
33414
+ "supports_vision": true,
33415
+ "supports_pdf_input": true
33416
+ },
33417
+ "standard/1024-x-1024/gpt-image-1.5": {
33418
+ "input_cost_per_image": 0.009,
33419
+ "litellm_provider": "openai",
33420
+ "mode": "image_generation",
33421
+ "supported_endpoints": [
33422
+ "/v1/images/generations",
33423
+ "/v1/images/edits"
33424
+ ],
33425
+ "supports_vision": true,
33426
+ "supports_pdf_input": true
33427
+ },
33428
+ "standard/1024-x-1536/gpt-image-1.5": {
33429
+ "input_cost_per_image": 0.013,
33430
+ "litellm_provider": "openai",
33431
+ "mode": "image_generation",
33432
+ "supported_endpoints": [
33433
+ "/v1/images/generations",
33434
+ "/v1/images/edits"
33435
+ ],
33436
+ "supports_vision": true,
33437
+ "supports_pdf_input": true
33438
+ },
33439
+ "standard/1536-x-1024/gpt-image-1.5": {
33440
+ "input_cost_per_image": 0.013,
33441
+ "litellm_provider": "openai",
33442
+ "mode": "image_generation",
33443
+ "supported_endpoints": [
33444
+ "/v1/images/generations",
33445
+ "/v1/images/edits"
33446
+ ],
33447
+ "supports_vision": true,
33448
+ "supports_pdf_input": true
33449
+ },
33450
+ "1024-x-1024/gpt-image-1.5": {
33451
+ "input_cost_per_image": 0.009,
33452
+ "litellm_provider": "openai",
33453
+ "mode": "image_generation",
33454
+ "supported_endpoints": [
33455
+ "/v1/images/generations",
33456
+ "/v1/images/edits"
33457
+ ],
33458
+ "supports_vision": true,
33459
+ "supports_pdf_input": true
33460
+ },
33461
+ "1024-x-1536/gpt-image-1.5": {
33462
+ "input_cost_per_image": 0.013,
33463
+ "litellm_provider": "openai",
33464
+ "mode": "image_generation",
33465
+ "supported_endpoints": [
33466
+ "/v1/images/generations",
33467
+ "/v1/images/edits"
33468
+ ],
33469
+ "supports_vision": true,
33470
+ "supports_pdf_input": true
33471
+ },
33472
+ "1536-x-1024/gpt-image-1.5": {
33473
+ "input_cost_per_image": 0.013,
33474
+ "litellm_provider": "openai",
33475
+ "mode": "image_generation",
33476
+ "supported_endpoints": [
33477
+ "/v1/images/generations",
33478
+ "/v1/images/edits"
33479
+ ],
33480
+ "supports_vision": true,
33481
+ "supports_pdf_input": true
33482
+ },
33483
+ "low/1024-x-1024/gpt-image-1.5-2025-12-16": {
33484
+ "input_cost_per_image": 0.009,
33485
+ "litellm_provider": "openai",
33486
+ "mode": "image_generation",
33487
+ "supported_endpoints": [
33488
+ "/v1/images/generations",
33489
+ "/v1/images/edits"
33490
+ ],
33491
+ "supports_vision": true,
33492
+ "supports_pdf_input": true
33493
+ },
33494
+ "low/1024-x-1536/gpt-image-1.5-2025-12-16": {
33495
+ "input_cost_per_image": 0.013,
33496
+ "litellm_provider": "openai",
33497
+ "mode": "image_generation",
33498
+ "supported_endpoints": [
33499
+ "/v1/images/generations",
33500
+ "/v1/images/edits"
33501
+ ],
33502
+ "supports_vision": true,
33503
+ "supports_pdf_input": true
33504
+ },
33505
+ "low/1536-x-1024/gpt-image-1.5-2025-12-16": {
33506
+ "input_cost_per_image": 0.013,
33507
+ "litellm_provider": "openai",
33508
+ "mode": "image_generation",
33509
+ "supported_endpoints": [
33510
+ "/v1/images/generations",
33511
+ "/v1/images/edits"
33512
+ ],
33513
+ "supports_vision": true,
33514
+ "supports_pdf_input": true
33515
+ },
33516
+ "medium/1024-x-1024/gpt-image-1.5-2025-12-16": {
33517
+ "input_cost_per_image": 0.034,
33518
+ "litellm_provider": "openai",
33519
+ "mode": "image_generation",
33520
+ "supported_endpoints": [
33521
+ "/v1/images/generations",
33522
+ "/v1/images/edits"
33523
+ ],
33524
+ "supports_vision": true,
33525
+ "supports_pdf_input": true
33526
+ },
33527
+ "medium/1024-x-1536/gpt-image-1.5-2025-12-16": {
33528
+ "input_cost_per_image": 0.05,
33529
+ "litellm_provider": "openai",
33530
+ "mode": "image_generation",
33531
+ "supported_endpoints": [
33532
+ "/v1/images/generations",
33533
+ "/v1/images/edits"
33534
+ ],
33535
+ "supports_vision": true,
33536
+ "supports_pdf_input": true
33537
+ },
33538
+ "medium/1536-x-1024/gpt-image-1.5-2025-12-16": {
33539
+ "input_cost_per_image": 0.05,
33540
+ "litellm_provider": "openai",
33541
+ "mode": "image_generation",
33542
+ "supported_endpoints": [
33543
+ "/v1/images/generations",
33544
+ "/v1/images/edits"
33545
+ ],
33546
+ "supports_vision": true,
33547
+ "supports_pdf_input": true
33548
+ },
33549
+ "high/1024-x-1024/gpt-image-1.5-2025-12-16": {
33550
+ "input_cost_per_image": 0.133,
33551
+ "litellm_provider": "openai",
33552
+ "mode": "image_generation",
33553
+ "supported_endpoints": [
33554
+ "/v1/images/generations",
33555
+ "/v1/images/edits"
33556
+ ],
33557
+ "supports_vision": true,
33558
+ "supports_pdf_input": true
33559
+ },
33560
+ "high/1024-x-1536/gpt-image-1.5-2025-12-16": {
33561
+ "input_cost_per_image": 0.2,
33562
+ "litellm_provider": "openai",
33563
+ "mode": "image_generation",
33564
+ "supported_endpoints": [
33565
+ "/v1/images/generations",
33566
+ "/v1/images/edits"
33567
+ ],
33568
+ "supports_vision": true,
33569
+ "supports_pdf_input": true
33570
+ },
33571
+ "high/1536-x-1024/gpt-image-1.5-2025-12-16": {
33572
+ "input_cost_per_image": 0.2,
33573
+ "litellm_provider": "openai",
33574
+ "mode": "image_generation",
33575
+ "supported_endpoints": [
33576
+ "/v1/images/generations",
33577
+ "/v1/images/edits"
33578
+ ],
33579
+ "supports_vision": true,
33580
+ "supports_pdf_input": true
33581
+ },
33582
+ "standard/1024-x-1024/gpt-image-1.5-2025-12-16": {
33583
+ "input_cost_per_image": 0.009,
33584
+ "litellm_provider": "openai",
33585
+ "mode": "image_generation",
33586
+ "supported_endpoints": [
33587
+ "/v1/images/generations",
33588
+ "/v1/images/edits"
33589
+ ],
33590
+ "supports_vision": true,
33591
+ "supports_pdf_input": true
33592
+ },
33593
+ "standard/1024-x-1536/gpt-image-1.5-2025-12-16": {
33594
+ "input_cost_per_image": 0.013,
33595
+ "litellm_provider": "openai",
33596
+ "mode": "image_generation",
33597
+ "supported_endpoints": [
33598
+ "/v1/images/generations",
33599
+ "/v1/images/edits"
33600
+ ],
33601
+ "supports_vision": true,
33602
+ "supports_pdf_input": true
33603
+ },
33604
+ "standard/1536-x-1024/gpt-image-1.5-2025-12-16": {
33605
+ "input_cost_per_image": 0.013,
33606
+ "litellm_provider": "openai",
33607
+ "mode": "image_generation",
33608
+ "supported_endpoints": [
33609
+ "/v1/images/generations",
33610
+ "/v1/images/edits"
33611
+ ],
33612
+ "supports_vision": true,
33613
+ "supports_pdf_input": true
33614
+ },
33615
+ "1024-x-1024/gpt-image-1.5-2025-12-16": {
33616
+ "input_cost_per_image": 0.009,
33617
+ "litellm_provider": "openai",
33618
+ "mode": "image_generation",
33619
+ "supported_endpoints": [
33620
+ "/v1/images/generations",
33621
+ "/v1/images/edits"
33622
+ ],
33623
+ "supports_vision": true,
33624
+ "supports_pdf_input": true
33625
+ },
33626
+ "1024-x-1536/gpt-image-1.5-2025-12-16": {
33627
+ "input_cost_per_image": 0.013,
33628
+ "litellm_provider": "openai",
33629
+ "mode": "image_generation",
33630
+ "supported_endpoints": [
33631
+ "/v1/images/generations",
33632
+ "/v1/images/edits"
33633
+ ],
33634
+ "supports_vision": true,
33635
+ "supports_pdf_input": true
33636
+ },
33637
+ "1536-x-1024/gpt-image-1.5-2025-12-16": {
33638
+ "input_cost_per_image": 0.013,
33639
+ "litellm_provider": "openai",
33640
+ "mode": "image_generation",
33641
+ "supported_endpoints": [
33642
+ "/v1/images/generations",
33643
+ "/v1/images/edits"
33644
+ ],
33645
+ "supports_vision": true,
33646
+ "supports_pdf_input": true
33647
+ },
33648
+ "zai/glm-4.7": {
33649
+ "cache_creation_input_token_cost": 0,
33650
+ "cache_read_input_token_cost": 1.1e-07,
33651
+ "input_cost_per_token": 6e-07,
33652
+ "output_cost_per_token": 2.2e-06,
33653
+ "litellm_provider": "zai",
33654
+ "max_input_tokens": 200000,
33655
+ "max_output_tokens": 128000,
33656
+ "mode": "chat",
33657
+ "supports_function_calling": true,
33658
+ "supports_reasoning": true,
33659
+ "supports_tool_choice": true,
33660
+ "source": "https://docs.z.ai/guides/overview/pricing"
33661
+ },
33662
+ "openrouter/google/gemini-3-flash-preview": {
33663
+ "cache_read_input_token_cost": 5e-08,
33664
+ "input_cost_per_audio_token": 1e-06,
33665
+ "input_cost_per_token": 5e-07,
33666
+ "litellm_provider": "openrouter",
33667
+ "max_audio_length_hours": 8.4,
33668
+ "max_audio_per_prompt": 1,
33669
+ "max_images_per_prompt": 3000,
33670
+ "max_input_tokens": 1048576,
33671
+ "max_output_tokens": 65535,
33672
+ "max_pdf_size_mb": 30,
33673
+ "max_tokens": 65535,
33674
+ "max_video_length": 1,
33675
+ "max_videos_per_prompt": 10,
33676
+ "mode": "chat",
33677
+ "output_cost_per_reasoning_token": 3e-06,
33678
+ "output_cost_per_token": 3e-06,
33679
+ "rpm": 2000,
33680
+ "source": "https://ai.google.dev/pricing/gemini-3",
33681
+ "supported_endpoints": [
33682
+ "/v1/chat/completions",
33683
+ "/v1/completions",
33684
+ "/v1/batch"
33685
+ ],
33686
+ "supported_modalities": [
33687
+ "text",
33688
+ "image",
33689
+ "audio",
33690
+ "video"
33691
+ ],
33692
+ "supported_output_modalities": [
33693
+ "text"
33694
+ ],
33695
+ "supports_audio_output": false,
33696
+ "supports_function_calling": true,
33697
+ "supports_parallel_function_calling": true,
33698
+ "supports_pdf_input": true,
33699
+ "supports_prompt_caching": true,
33700
+ "supports_reasoning": true,
33701
+ "supports_response_schema": true,
33702
+ "supports_system_messages": true,
33703
+ "supports_tool_choice": true,
33704
+ "supports_url_context": true,
33705
+ "supports_vision": true,
33706
+ "supports_web_search": true,
33707
+ "tpm": 800000
33708
+ },
33709
+ "amazon.nova-2-multimodal-embeddings-v1:0": {
33710
+ "litellm_provider": "bedrock",
33711
+ "max_input_tokens": 8172,
33712
+ "max_tokens": 8172,
33713
+ "mode": "embedding",
33714
+ "input_cost_per_token": 1.35e-07,
33715
+ "input_cost_per_image": 6e-05,
33716
+ "input_cost_per_video_per_second": 0.0007,
33717
+ "input_cost_per_audio_per_second": 0.00014,
33718
+ "output_cost_per_token": 0.0,
33719
+ "output_vector_size": 3072,
33720
+ "source": "https://us-east-1.console.aws.amazon.com/bedrock/home?region=us-east-1#/model-catalog/serverless/amazon.nova-2-multimodal-embeddings-v1:0",
33721
+ "supports_embedding_image_input": true,
33722
+ "supports_image_input": true,
33723
+ "supports_video_input": true,
33724
+ "supports_audio_input": true
33725
+ },
33726
+ "gigachat/GigaChat-2-Lite": {
33727
+ "input_cost_per_token": 0.0,
33728
+ "litellm_provider": "gigachat",
33729
+ "max_input_tokens": 128000,
33730
+ "max_output_tokens": 8192,
33731
+ "max_tokens": 8192,
33732
+ "mode": "chat",
33733
+ "output_cost_per_token": 0.0,
33734
+ "supports_function_calling": true,
33735
+ "supports_system_messages": true
33736
+ },
33737
+ "gigachat/GigaChat-2-Max": {
33738
+ "input_cost_per_token": 0.0,
33739
+ "litellm_provider": "gigachat",
33740
+ "max_input_tokens": 128000,
33741
+ "max_output_tokens": 8192,
33742
+ "max_tokens": 8192,
33743
+ "mode": "chat",
33744
+ "output_cost_per_token": 0.0,
33745
+ "supports_function_calling": true,
33746
+ "supports_system_messages": true,
33747
+ "supports_vision": true
33748
+ },
33749
+ "gigachat/GigaChat-2-Pro": {
33750
+ "input_cost_per_token": 0.0,
33751
+ "litellm_provider": "gigachat",
33752
+ "max_input_tokens": 128000,
33753
+ "max_output_tokens": 8192,
33754
+ "max_tokens": 8192,
33755
+ "mode": "chat",
33756
+ "output_cost_per_token": 0.0,
33757
+ "supports_function_calling": true,
33758
+ "supports_system_messages": true,
33759
+ "supports_vision": true
33760
+ },
33761
+ "gigachat/Embeddings": {
33762
+ "input_cost_per_token": 0.0,
33763
+ "litellm_provider": "gigachat",
33764
+ "max_input_tokens": 512,
33765
+ "max_tokens": 512,
33766
+ "mode": "embedding",
33767
+ "output_cost_per_token": 0.0,
33768
+ "output_vector_size": 1024
33769
+ },
33770
+ "gigachat/Embeddings-2": {
33771
+ "input_cost_per_token": 0.0,
33772
+ "litellm_provider": "gigachat",
33773
+ "max_input_tokens": 512,
33774
+ "max_tokens": 512,
33775
+ "mode": "embedding",
33776
+ "output_cost_per_token": 0.0,
33777
+ "output_vector_size": 1024
33778
+ },
33779
+ "gigachat/EmbeddingsGigaR": {
33780
+ "input_cost_per_token": 0.0,
33781
+ "litellm_provider": "gigachat",
33782
+ "max_input_tokens": 4096,
33783
+ "max_tokens": 4096,
33784
+ "mode": "embedding",
33785
+ "output_cost_per_token": 0.0,
33786
+ "output_vector_size": 2560
33787
+ },
33788
+ "llamagate/llama-3.1-8b": {
33789
+ "max_tokens": 8192,
33790
+ "max_input_tokens": 131072,
33791
+ "max_output_tokens": 8192,
33792
+ "input_cost_per_token": 3e-08,
33793
+ "output_cost_per_token": 5e-08,
33794
+ "litellm_provider": "llamagate",
33795
+ "mode": "chat",
33796
+ "supports_function_calling": true,
33797
+ "supports_response_schema": true
33798
+ },
33799
+ "llamagate/llama-3.2-3b": {
33800
+ "max_tokens": 8192,
33801
+ "max_input_tokens": 131072,
33802
+ "max_output_tokens": 8192,
33803
+ "input_cost_per_token": 4e-08,
33804
+ "output_cost_per_token": 8e-08,
33805
+ "litellm_provider": "llamagate",
33806
+ "mode": "chat",
33807
+ "supports_function_calling": true,
33808
+ "supports_response_schema": true
33809
+ },
33810
+ "llamagate/mistral-7b-v0.3": {
33811
+ "max_tokens": 8192,
33812
+ "max_input_tokens": 32768,
33813
+ "max_output_tokens": 8192,
33814
+ "input_cost_per_token": 1e-07,
33815
+ "output_cost_per_token": 1.5e-07,
33816
+ "litellm_provider": "llamagate",
33817
+ "mode": "chat",
33818
+ "supports_function_calling": true,
33819
+ "supports_response_schema": true
33820
+ },
33821
+ "llamagate/qwen3-8b": {
33822
+ "max_tokens": 8192,
33823
+ "max_input_tokens": 32768,
33824
+ "max_output_tokens": 8192,
33825
+ "input_cost_per_token": 4e-08,
33826
+ "output_cost_per_token": 1.4e-07,
33827
+ "litellm_provider": "llamagate",
33828
+ "mode": "chat",
33829
+ "supports_function_calling": true,
33830
+ "supports_response_schema": true
33831
+ },
33832
+ "llamagate/dolphin3-8b": {
33833
+ "max_tokens": 8192,
33834
+ "max_input_tokens": 128000,
33835
+ "max_output_tokens": 8192,
33836
+ "input_cost_per_token": 8e-08,
33837
+ "output_cost_per_token": 1.5e-07,
33838
+ "litellm_provider": "llamagate",
33839
+ "mode": "chat",
33840
+ "supports_function_calling": true,
33841
+ "supports_response_schema": true
33842
+ },
33843
+ "llamagate/deepseek-r1-8b": {
33844
+ "max_tokens": 16384,
33845
+ "max_input_tokens": 65536,
33846
+ "max_output_tokens": 16384,
33847
+ "input_cost_per_token": 1e-07,
33848
+ "output_cost_per_token": 2e-07,
33849
+ "litellm_provider": "llamagate",
33850
+ "mode": "chat",
33851
+ "supports_function_calling": true,
33852
+ "supports_response_schema": true,
33853
+ "supports_reasoning": true
33854
+ },
33855
+ "llamagate/deepseek-r1-7b-qwen": {
33856
+ "max_tokens": 16384,
33857
+ "max_input_tokens": 131072,
33858
+ "max_output_tokens": 16384,
33859
+ "input_cost_per_token": 8e-08,
33860
+ "output_cost_per_token": 1.5e-07,
33861
+ "litellm_provider": "llamagate",
33862
+ "mode": "chat",
33863
+ "supports_function_calling": true,
33864
+ "supports_response_schema": true,
33865
+ "supports_reasoning": true
33866
+ },
33867
+ "llamagate/openthinker-7b": {
33868
+ "max_tokens": 8192,
33869
+ "max_input_tokens": 32768,
33870
+ "max_output_tokens": 8192,
33871
+ "input_cost_per_token": 8e-08,
33872
+ "output_cost_per_token": 1.5e-07,
33873
+ "litellm_provider": "llamagate",
33874
+ "mode": "chat",
33875
+ "supports_function_calling": true,
33876
+ "supports_response_schema": true,
33877
+ "supports_reasoning": true
33878
+ },
33879
+ "llamagate/qwen2.5-coder-7b": {
33880
+ "max_tokens": 8192,
33881
+ "max_input_tokens": 32768,
33882
+ "max_output_tokens": 8192,
33883
+ "input_cost_per_token": 6e-08,
33884
+ "output_cost_per_token": 1.2e-07,
33885
+ "litellm_provider": "llamagate",
33886
+ "mode": "chat",
33887
+ "supports_function_calling": true,
33888
+ "supports_response_schema": true
33889
+ },
33890
+ "llamagate/deepseek-coder-6.7b": {
33891
+ "max_tokens": 4096,
33892
+ "max_input_tokens": 16384,
33893
+ "max_output_tokens": 4096,
33894
+ "input_cost_per_token": 6e-08,
33895
+ "output_cost_per_token": 1.2e-07,
33896
+ "litellm_provider": "llamagate",
33897
+ "mode": "chat",
33898
+ "supports_function_calling": true,
33899
+ "supports_response_schema": true
33900
+ },
33901
+ "llamagate/codellama-7b": {
33902
+ "max_tokens": 4096,
33903
+ "max_input_tokens": 16384,
33904
+ "max_output_tokens": 4096,
33905
+ "input_cost_per_token": 6e-08,
33906
+ "output_cost_per_token": 1.2e-07,
33907
+ "litellm_provider": "llamagate",
33908
+ "mode": "chat",
33909
+ "supports_function_calling": true,
33910
+ "supports_response_schema": true
33911
+ },
33912
+ "llamagate/qwen3-vl-8b": {
33913
+ "max_tokens": 8192,
33914
+ "max_input_tokens": 32768,
33915
+ "max_output_tokens": 8192,
33916
+ "input_cost_per_token": 1.5e-07,
33917
+ "output_cost_per_token": 5.5e-07,
33918
+ "litellm_provider": "llamagate",
33919
+ "mode": "chat",
33920
+ "supports_function_calling": true,
33921
+ "supports_response_schema": true,
33922
+ "supports_vision": true
33923
+ },
33924
+ "llamagate/llava-7b": {
33925
+ "max_tokens": 2048,
33926
+ "max_input_tokens": 4096,
33927
+ "max_output_tokens": 2048,
33928
+ "input_cost_per_token": 1e-07,
33929
+ "output_cost_per_token": 2e-07,
33930
+ "litellm_provider": "llamagate",
33931
+ "mode": "chat",
33932
+ "supports_response_schema": true,
33933
+ "supports_vision": true
33934
+ },
33935
+ "llamagate/gemma3-4b": {
33936
+ "max_tokens": 8192,
33937
+ "max_input_tokens": 128000,
33938
+ "max_output_tokens": 8192,
33939
+ "input_cost_per_token": 3e-08,
33940
+ "output_cost_per_token": 8e-08,
33941
+ "litellm_provider": "llamagate",
33942
+ "mode": "chat",
33943
+ "supports_function_calling": true,
33944
+ "supports_response_schema": true,
33945
+ "supports_vision": true
33946
+ },
33947
+ "llamagate/nomic-embed-text": {
33948
+ "max_tokens": 8192,
33949
+ "max_input_tokens": 8192,
33950
+ "input_cost_per_token": 2e-08,
33951
+ "output_cost_per_token": 0,
33952
+ "litellm_provider": "llamagate",
33953
+ "mode": "embedding"
33954
+ },
33955
+ "llamagate/qwen3-embedding-8b": {
33956
+ "max_tokens": 40960,
33957
+ "max_input_tokens": 40960,
33958
+ "input_cost_per_token": 2e-08,
33959
+ "output_cost_per_token": 0,
33960
+ "litellm_provider": "llamagate",
33961
+ "mode": "embedding"
33962
+ },
33963
+ "azure_ai/flux.2-pro": {
33964
+ "litellm_provider": "azure_ai",
33965
+ "mode": "image_generation",
33966
+ "output_cost_per_image": 0.04,
33967
+ "source": "https://ai.azure.com/explore/models/flux.2-pro/version/1/registry/azureml-blackforestlabs",
33968
+ "supported_endpoints": [
33969
+ "/v1/images/generations"
33970
+ ]
33971
+ },
33972
+ "vertex_ai/zai-org/glm-4.7-maas": {
33973
+ "input_cost_per_token": 3e-07,
33974
+ "litellm_provider": "vertex_ai-zai_models",
33975
+ "max_input_tokens": 200000,
33976
+ "max_output_tokens": 128000,
33977
+ "max_tokens": 128000,
33978
+ "mode": "chat",
33979
+ "output_cost_per_token": 1.2e-06,
33980
+ "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing#partner-models",
33981
+ "supports_function_calling": true,
33982
+ "supports_reasoning": true,
33983
+ "supports_tool_choice": true
33984
+ },
33985
+ "novita/deepseek/deepseek-v3.2": {
33986
+ "litellm_provider": "novita",
33987
+ "mode": "chat",
33988
+ "input_cost_per_token": 0.00269,
33989
+ "output_cost_per_token": 0.004,
33990
+ "max_input_tokens": 163840,
33991
+ "max_output_tokens": 65536,
33992
+ "max_tokens": 65536,
33993
+ "supports_function_calling": true,
33994
+ "supports_parallel_function_calling": true,
33995
+ "supports_tool_choice": true,
33996
+ "supports_system_messages": true,
33997
+ "supports_response_schema": true,
33998
+ "cache_read_input_token_cost": 0.001345,
33999
+ "input_cost_per_token_cache_hit": 0.001345,
34000
+ "supports_reasoning": true
34001
+ },
34002
+ "novita/minimax/minimax-m2.1": {
34003
+ "litellm_provider": "novita",
34004
+ "mode": "chat",
34005
+ "input_cost_per_token": 0.003,
34006
+ "output_cost_per_token": 0.012,
34007
+ "max_input_tokens": 204800,
34008
+ "max_output_tokens": 131072,
34009
+ "max_tokens": 131072,
34010
+ "supports_function_calling": true,
34011
+ "supports_parallel_function_calling": true,
34012
+ "supports_tool_choice": true,
34013
+ "supports_system_messages": true,
34014
+ "supports_response_schema": true,
34015
+ "cache_read_input_token_cost": 0.0003,
34016
+ "input_cost_per_token_cache_hit": 0.0003,
34017
+ "supports_reasoning": true
34018
+ },
34019
+ "novita/zai-org/glm-4.7": {
34020
+ "litellm_provider": "novita",
34021
+ "mode": "chat",
34022
+ "input_cost_per_token": 0.006,
34023
+ "output_cost_per_token": 0.022,
34024
+ "max_input_tokens": 204800,
34025
+ "max_output_tokens": 131072,
34026
+ "max_tokens": 131072,
34027
+ "supports_function_calling": true,
34028
+ "supports_parallel_function_calling": true,
34029
+ "supports_tool_choice": true,
34030
+ "supports_system_messages": true,
34031
+ "supports_response_schema": true,
34032
+ "cache_read_input_token_cost": 0.0011,
34033
+ "input_cost_per_token_cache_hit": 0.0011,
34034
+ "supports_reasoning": true
34035
+ },
34036
+ "novita/xiaomimimo/mimo-v2-flash": {
34037
+ "litellm_provider": "novita",
34038
+ "mode": "chat",
34039
+ "input_cost_per_token": 0.001,
34040
+ "output_cost_per_token": 0.003,
34041
+ "max_input_tokens": 262144,
34042
+ "max_output_tokens": 32000,
34043
+ "max_tokens": 32000,
34044
+ "supports_function_calling": true,
34045
+ "supports_parallel_function_calling": true,
34046
+ "supports_tool_choice": true,
34047
+ "supports_system_messages": true,
34048
+ "supports_response_schema": true,
34049
+ "cache_read_input_token_cost": 0.0002,
34050
+ "input_cost_per_token_cache_hit": 0.0002,
34051
+ "supports_reasoning": true
34052
+ },
34053
+ "novita/zai-org/autoglm-phone-9b-multilingual": {
34054
+ "litellm_provider": "novita",
34055
+ "mode": "chat",
34056
+ "input_cost_per_token": 0.00035,
34057
+ "output_cost_per_token": 0.00138,
34058
+ "max_input_tokens": 65536,
34059
+ "max_output_tokens": 65536,
34060
+ "max_tokens": 65536,
34061
+ "supports_vision": true,
34062
+ "supports_system_messages": true
34063
+ },
34064
+ "novita/moonshotai/kimi-k2-thinking": {
34065
+ "litellm_provider": "novita",
34066
+ "mode": "chat",
34067
+ "input_cost_per_token": 0.0048,
34068
+ "output_cost_per_token": 0.02,
34069
+ "max_input_tokens": 262144,
34070
+ "max_output_tokens": 262144,
34071
+ "max_tokens": 262144,
34072
+ "supports_function_calling": true,
34073
+ "supports_parallel_function_calling": true,
34074
+ "supports_tool_choice": true,
34075
+ "supports_system_messages": true,
34076
+ "supports_response_schema": true,
34077
+ "supports_reasoning": true
34078
+ },
34079
+ "novita/minimax/minimax-m2": {
34080
+ "litellm_provider": "novita",
34081
+ "mode": "chat",
34082
+ "input_cost_per_token": 0.0024,
34083
+ "output_cost_per_token": 0.0096,
34084
+ "max_input_tokens": 204800,
34085
+ "max_output_tokens": 131072,
34086
+ "max_tokens": 131072,
34087
+ "supports_function_calling": true,
34088
+ "supports_parallel_function_calling": true,
34089
+ "supports_tool_choice": true,
34090
+ "supports_system_messages": true,
34091
+ "cache_read_input_token_cost": 0.00024,
34092
+ "input_cost_per_token_cache_hit": 0.00024,
34093
+ "supports_reasoning": true
34094
+ },
34095
+ "novita/paddlepaddle/paddleocr-vl": {
34096
+ "litellm_provider": "novita",
34097
+ "mode": "chat",
34098
+ "input_cost_per_token": 0.00016,
34099
+ "output_cost_per_token": 0.00016,
34100
+ "max_input_tokens": 16384,
34101
+ "max_output_tokens": 16384,
34102
+ "max_tokens": 16384,
34103
+ "supports_vision": true,
34104
+ "supports_system_messages": true
34105
+ },
34106
+ "novita/deepseek/deepseek-v3.2-exp": {
34107
+ "litellm_provider": "novita",
34108
+ "mode": "chat",
34109
+ "input_cost_per_token": 0.00216,
34110
+ "output_cost_per_token": 0.00328,
34111
+ "max_input_tokens": 163840,
34112
+ "max_output_tokens": 65536,
34113
+ "max_tokens": 65536,
34114
+ "supports_function_calling": true,
34115
+ "supports_parallel_function_calling": true,
34116
+ "supports_tool_choice": true,
34117
+ "supports_system_messages": true,
34118
+ "supports_response_schema": true,
34119
+ "supports_reasoning": true
34120
+ },
34121
+ "novita/qwen/qwen3-vl-235b-a22b-thinking": {
34122
+ "litellm_provider": "novita",
34123
+ "mode": "chat",
34124
+ "input_cost_per_token": 0.00784,
34125
+ "output_cost_per_token": 0.0316,
34126
+ "max_input_tokens": 131072,
34127
+ "max_output_tokens": 32768,
34128
+ "max_tokens": 32768,
34129
+ "supports_vision": true,
34130
+ "supports_system_messages": true,
34131
+ "supports_reasoning": true
34132
+ },
34133
+ "novita/zai-org/glm-4.6v": {
34134
+ "litellm_provider": "novita",
34135
+ "mode": "chat",
34136
+ "input_cost_per_token": 0.003,
34137
+ "output_cost_per_token": 0.009,
34138
+ "max_input_tokens": 131072,
34139
+ "max_output_tokens": 32768,
34140
+ "max_tokens": 32768,
34141
+ "supports_function_calling": true,
34142
+ "supports_parallel_function_calling": true,
34143
+ "supports_tool_choice": true,
34144
+ "supports_vision": true,
34145
+ "supports_system_messages": true,
34146
+ "supports_response_schema": true,
34147
+ "cache_read_input_token_cost": 0.00055,
34148
+ "input_cost_per_token_cache_hit": 0.00055,
34149
+ "supports_reasoning": true
34150
+ },
34151
+ "novita/zai-org/glm-4.6": {
34152
+ "litellm_provider": "novita",
34153
+ "mode": "chat",
34154
+ "input_cost_per_token": 0.0044,
34155
+ "output_cost_per_token": 0.0176,
34156
+ "max_input_tokens": 204800,
34157
+ "max_output_tokens": 131072,
34158
+ "max_tokens": 131072,
34159
+ "supports_function_calling": true,
34160
+ "supports_parallel_function_calling": true,
34161
+ "supports_tool_choice": true,
34162
+ "supports_system_messages": true,
34163
+ "supports_response_schema": true,
34164
+ "cache_read_input_token_cost": 0.00088,
34165
+ "input_cost_per_token_cache_hit": 0.00088,
34166
+ "supports_reasoning": true
34167
+ },
34168
+ "novita/qwen/qwen3-next-80b-a3b-instruct": {
34169
+ "litellm_provider": "novita",
34170
+ "mode": "chat",
34171
+ "input_cost_per_token": 0.0012,
34172
+ "output_cost_per_token": 0.012,
34173
+ "max_input_tokens": 131072,
34174
+ "max_output_tokens": 32768,
34175
+ "max_tokens": 32768,
34176
+ "supports_function_calling": true,
34177
+ "supports_parallel_function_calling": true,
34178
+ "supports_tool_choice": true,
34179
+ "supports_system_messages": true,
34180
+ "supports_response_schema": true
34181
+ },
34182
+ "novita/qwen/qwen3-next-80b-a3b-thinking": {
34183
+ "litellm_provider": "novita",
34184
+ "mode": "chat",
34185
+ "input_cost_per_token": 0.0012,
34186
+ "output_cost_per_token": 0.012,
34187
+ "max_input_tokens": 131072,
34188
+ "max_output_tokens": 32768,
34189
+ "max_tokens": 32768,
34190
+ "supports_function_calling": true,
34191
+ "supports_parallel_function_calling": true,
34192
+ "supports_tool_choice": true,
34193
+ "supports_system_messages": true,
34194
+ "supports_response_schema": true,
34195
+ "supports_reasoning": true
34196
+ },
34197
+ "novita/deepseek/deepseek-ocr": {
34198
+ "litellm_provider": "novita",
34199
+ "mode": "chat",
34200
+ "input_cost_per_token": 0.00024,
34201
+ "output_cost_per_token": 0.00024,
34202
+ "max_input_tokens": 8192,
34203
+ "max_output_tokens": 8192,
34204
+ "max_tokens": 8192,
34205
+ "supports_tool_choice": true,
34206
+ "supports_vision": true,
34207
+ "supports_system_messages": true,
34208
+ "supports_response_schema": true
34209
+ },
34210
+ "novita/deepseek/deepseek-v3.1-terminus": {
34211
+ "litellm_provider": "novita",
34212
+ "mode": "chat",
34213
+ "input_cost_per_token": 0.00216,
34214
+ "output_cost_per_token": 0.008,
34215
+ "max_input_tokens": 131072,
34216
+ "max_output_tokens": 32768,
34217
+ "max_tokens": 32768,
34218
+ "supports_function_calling": true,
34219
+ "supports_parallel_function_calling": true,
34220
+ "supports_tool_choice": true,
34221
+ "supports_system_messages": true,
34222
+ "supports_response_schema": true,
34223
+ "cache_read_input_token_cost": 0.00108,
34224
+ "input_cost_per_token_cache_hit": 0.00108,
34225
+ "supports_reasoning": true
34226
+ },
34227
+ "novita/qwen/qwen3-vl-235b-a22b-instruct": {
34228
+ "litellm_provider": "novita",
34229
+ "mode": "chat",
34230
+ "input_cost_per_token": 0.0024,
34231
+ "output_cost_per_token": 0.012,
34232
+ "max_input_tokens": 131072,
34233
+ "max_output_tokens": 32768,
34234
+ "max_tokens": 32768,
34235
+ "supports_function_calling": true,
34236
+ "supports_parallel_function_calling": true,
34237
+ "supports_tool_choice": true,
34238
+ "supports_vision": true,
34239
+ "supports_system_messages": true,
34240
+ "supports_response_schema": true
34241
+ },
34242
+ "novita/qwen/qwen3-max": {
34243
+ "litellm_provider": "novita",
34244
+ "mode": "chat",
34245
+ "input_cost_per_token": 0.01688,
34246
+ "output_cost_per_token": 0.0676,
34247
+ "max_input_tokens": 262144,
34248
+ "max_output_tokens": 65536,
34249
+ "max_tokens": 65536,
34250
+ "supports_function_calling": true,
34251
+ "supports_parallel_function_calling": true,
34252
+ "supports_tool_choice": true,
34253
+ "supports_system_messages": true,
34254
+ "supports_response_schema": true
34255
+ },
34256
+ "novita/skywork/r1v4-lite": {
34257
+ "litellm_provider": "novita",
34258
+ "mode": "chat",
34259
+ "input_cost_per_token": 0.002,
34260
+ "output_cost_per_token": 0.006,
34261
+ "max_input_tokens": 262144,
34262
+ "max_output_tokens": 65536,
34263
+ "max_tokens": 65536,
34264
+ "supports_tool_choice": true,
34265
+ "supports_vision": true,
34266
+ "supports_system_messages": true,
34267
+ "supports_response_schema": true
34268
+ },
34269
+ "novita/deepseek/deepseek-v3.1": {
34270
+ "litellm_provider": "novita",
34271
+ "mode": "chat",
34272
+ "input_cost_per_token": 0.00216,
34273
+ "output_cost_per_token": 0.008,
34274
+ "max_input_tokens": 131072,
34275
+ "max_output_tokens": 32768,
34276
+ "max_tokens": 32768,
34277
+ "supports_function_calling": true,
34278
+ "supports_parallel_function_calling": true,
34279
+ "supports_tool_choice": true,
34280
+ "supports_system_messages": true,
34281
+ "supports_response_schema": true,
34282
+ "cache_read_input_token_cost": 0.00108,
34283
+ "input_cost_per_token_cache_hit": 0.00108,
34284
+ "supports_reasoning": true
34285
+ },
34286
+ "novita/moonshotai/kimi-k2-0905": {
34287
+ "litellm_provider": "novita",
34288
+ "mode": "chat",
34289
+ "input_cost_per_token": 0.0048,
34290
+ "output_cost_per_token": 0.02,
34291
+ "max_input_tokens": 262144,
34292
+ "max_output_tokens": 262144,
34293
+ "max_tokens": 262144,
34294
+ "supports_function_calling": true,
34295
+ "supports_parallel_function_calling": true,
34296
+ "supports_tool_choice": true,
34297
+ "supports_system_messages": true,
34298
+ "supports_response_schema": true
34299
+ },
34300
+ "novita/qwen/qwen3-coder-480b-a35b-instruct": {
34301
+ "litellm_provider": "novita",
34302
+ "mode": "chat",
34303
+ "input_cost_per_token": 0.0024,
34304
+ "output_cost_per_token": 0.0104,
34305
+ "max_input_tokens": 262144,
34306
+ "max_output_tokens": 65536,
34307
+ "max_tokens": 65536,
34308
+ "supports_function_calling": true,
34309
+ "supports_parallel_function_calling": true,
34310
+ "supports_tool_choice": true,
34311
+ "supports_system_messages": true,
34312
+ "supports_response_schema": true
34313
+ },
34314
+ "novita/qwen/qwen3-coder-30b-a3b-instruct": {
34315
+ "litellm_provider": "novita",
34316
+ "mode": "chat",
34317
+ "input_cost_per_token": 0.0007,
34318
+ "output_cost_per_token": 0.0027,
34319
+ "max_input_tokens": 160000,
34320
+ "max_output_tokens": 32768,
34321
+ "max_tokens": 32768,
34322
+ "supports_function_calling": true,
34323
+ "supports_parallel_function_calling": true,
34324
+ "supports_tool_choice": true,
34325
+ "supports_system_messages": true,
34326
+ "supports_response_schema": true
34327
+ },
34328
+ "novita/openai/gpt-oss-120b": {
34329
+ "litellm_provider": "novita",
34330
+ "mode": "chat",
34331
+ "input_cost_per_token": 0.0004,
34332
+ "output_cost_per_token": 0.002,
34333
+ "max_input_tokens": 131072,
34334
+ "max_output_tokens": 32768,
34335
+ "max_tokens": 32768,
34336
+ "supports_function_calling": true,
34337
+ "supports_parallel_function_calling": true,
34338
+ "supports_tool_choice": true,
34339
+ "supports_vision": true,
34340
+ "supports_system_messages": true,
34341
+ "supports_response_schema": true,
34342
+ "supports_reasoning": true
34343
+ },
34344
+ "novita/moonshotai/kimi-k2-instruct": {
34345
+ "litellm_provider": "novita",
34346
+ "mode": "chat",
34347
+ "input_cost_per_token": 0.00456,
34348
+ "output_cost_per_token": 0.0184,
34349
+ "max_input_tokens": 131072,
34350
+ "max_output_tokens": 131072,
34351
+ "max_tokens": 131072,
34352
+ "supports_function_calling": true,
34353
+ "supports_parallel_function_calling": true,
34354
+ "supports_tool_choice": true,
34355
+ "supports_system_messages": true,
34356
+ "supports_response_schema": true
34357
+ },
34358
+ "novita/deepseek/deepseek-v3-0324": {
34359
+ "litellm_provider": "novita",
34360
+ "mode": "chat",
34361
+ "input_cost_per_token": 0.00216,
34362
+ "output_cost_per_token": 0.00896,
34363
+ "max_input_tokens": 163840,
34364
+ "max_output_tokens": 163840,
34365
+ "max_tokens": 163840,
34366
+ "supports_function_calling": true,
34367
+ "supports_parallel_function_calling": true,
34368
+ "supports_tool_choice": true,
34369
+ "supports_system_messages": true,
34370
+ "supports_response_schema": true,
34371
+ "cache_read_input_token_cost": 0.00108,
34372
+ "input_cost_per_token_cache_hit": 0.00108
34373
+ },
34374
+ "novita/zai-org/glm-4.5": {
34375
+ "litellm_provider": "novita",
34376
+ "mode": "chat",
34377
+ "input_cost_per_token": 0.0048,
34378
+ "output_cost_per_token": 0.0176,
34379
+ "max_input_tokens": 131072,
34380
+ "max_output_tokens": 98304,
34381
+ "max_tokens": 98304,
34382
+ "supports_function_calling": true,
34383
+ "supports_parallel_function_calling": true,
34384
+ "supports_tool_choice": true,
34385
+ "supports_system_messages": true,
34386
+ "cache_read_input_token_cost": 0.00088,
34387
+ "input_cost_per_token_cache_hit": 0.00088,
34388
+ "supports_reasoning": true
34389
+ },
34390
+ "novita/qwen/qwen3-235b-a22b-thinking-2507": {
34391
+ "litellm_provider": "novita",
34392
+ "mode": "chat",
34393
+ "input_cost_per_token": 0.0024,
34394
+ "output_cost_per_token": 0.024,
34395
+ "max_input_tokens": 131072,
34396
+ "max_output_tokens": 32768,
34397
+ "max_tokens": 32768,
34398
+ "supports_function_calling": true,
34399
+ "supports_parallel_function_calling": true,
34400
+ "supports_tool_choice": true,
34401
+ "supports_system_messages": true,
34402
+ "supports_reasoning": true
34403
+ },
34404
+ "novita/meta-llama/llama-3.1-8b-instruct": {
34405
+ "litellm_provider": "novita",
34406
+ "mode": "chat",
34407
+ "input_cost_per_token": 0.0002,
34408
+ "output_cost_per_token": 0.0005,
34409
+ "max_input_tokens": 16384,
34410
+ "max_output_tokens": 16384,
34411
+ "max_tokens": 16384,
34412
+ "supports_system_messages": true
34413
+ },
34414
+ "novita/google/gemma-3-12b-it": {
34415
+ "litellm_provider": "novita",
34416
+ "mode": "chat",
34417
+ "input_cost_per_token": 0.0004,
34418
+ "output_cost_per_token": 0.0008,
34419
+ "max_input_tokens": 131072,
34420
+ "max_output_tokens": 8192,
34421
+ "max_tokens": 8192,
34422
+ "supports_tool_choice": true,
34423
+ "supports_vision": true,
34424
+ "supports_system_messages": true,
34425
+ "supports_response_schema": true
34426
+ },
34427
+ "novita/zai-org/glm-4.5v": {
34428
+ "litellm_provider": "novita",
34429
+ "mode": "chat",
34430
+ "input_cost_per_token": 0.0048,
34431
+ "output_cost_per_token": 0.0144,
34432
+ "max_input_tokens": 65536,
34433
+ "max_output_tokens": 16384,
34434
+ "max_tokens": 16384,
34435
+ "supports_function_calling": true,
34436
+ "supports_parallel_function_calling": true,
34437
+ "supports_tool_choice": true,
34438
+ "supports_vision": true,
34439
+ "supports_system_messages": true,
34440
+ "supports_response_schema": true,
34441
+ "cache_read_input_token_cost": 0.00088,
34442
+ "input_cost_per_token_cache_hit": 0.00088,
34443
+ "supports_reasoning": true
34444
+ },
34445
+ "novita/openai/gpt-oss-20b": {
34446
+ "litellm_provider": "novita",
34447
+ "mode": "chat",
34448
+ "input_cost_per_token": 0.00032,
34449
+ "output_cost_per_token": 0.0012,
34450
+ "max_input_tokens": 131072,
34451
+ "max_output_tokens": 32768,
34452
+ "max_tokens": 32768,
34453
+ "supports_tool_choice": true,
34454
+ "supports_vision": true,
34455
+ "supports_system_messages": true,
34456
+ "supports_response_schema": true,
34457
+ "supports_reasoning": true
34458
+ },
34459
+ "novita/qwen/qwen3-235b-a22b-instruct-2507": {
34460
+ "litellm_provider": "novita",
34461
+ "mode": "chat",
34462
+ "input_cost_per_token": 0.00072,
34463
+ "output_cost_per_token": 0.00464,
34464
+ "max_input_tokens": 131072,
34465
+ "max_output_tokens": 16384,
34466
+ "max_tokens": 16384,
34467
+ "supports_function_calling": true,
34468
+ "supports_parallel_function_calling": true,
34469
+ "supports_tool_choice": true,
34470
+ "supports_system_messages": true,
34471
+ "supports_response_schema": true
34472
+ },
34473
+ "novita/deepseek/deepseek-r1-distill-qwen-14b": {
34474
+ "litellm_provider": "novita",
34475
+ "mode": "chat",
34476
+ "input_cost_per_token": 0.0012,
34477
+ "output_cost_per_token": 0.0012,
34478
+ "max_input_tokens": 32768,
34479
+ "max_output_tokens": 16384,
34480
+ "max_tokens": 16384,
34481
+ "supports_tool_choice": true,
34482
+ "supports_system_messages": true,
34483
+ "supports_response_schema": true,
34484
+ "supports_reasoning": true
34485
+ },
34486
+ "novita/meta-llama/llama-3.3-70b-instruct": {
34487
+ "litellm_provider": "novita",
34488
+ "mode": "chat",
34489
+ "input_cost_per_token": 0.00108,
34490
+ "output_cost_per_token": 0.0032,
34491
+ "max_input_tokens": 131072,
34492
+ "max_output_tokens": 120000,
34493
+ "max_tokens": 120000,
34494
+ "supports_function_calling": true,
34495
+ "supports_parallel_function_calling": true,
34496
+ "supports_tool_choice": true,
34497
+ "supports_system_messages": true
34498
+ },
34499
+ "novita/qwen/qwen-2.5-72b-instruct": {
34500
+ "litellm_provider": "novita",
34501
+ "mode": "chat",
34502
+ "input_cost_per_token": 0.00304,
34503
+ "output_cost_per_token": 0.0032,
34504
+ "max_input_tokens": 32000,
34505
+ "max_output_tokens": 8192,
34506
+ "max_tokens": 8192,
34507
+ "supports_function_calling": true,
34508
+ "supports_parallel_function_calling": true,
34509
+ "supports_tool_choice": true,
34510
+ "supports_system_messages": true,
34511
+ "supports_response_schema": true
34512
+ },
34513
+ "novita/mistralai/mistral-nemo": {
34514
+ "litellm_provider": "novita",
34515
+ "mode": "chat",
34516
+ "input_cost_per_token": 0.00032,
34517
+ "output_cost_per_token": 0.00136,
34518
+ "max_input_tokens": 60288,
34519
+ "max_output_tokens": 16000,
34520
+ "max_tokens": 16000,
34521
+ "supports_tool_choice": true,
34522
+ "supports_system_messages": true,
34523
+ "supports_response_schema": true
34524
+ },
34525
+ "novita/minimaxai/minimax-m1-80k": {
34526
+ "litellm_provider": "novita",
34527
+ "mode": "chat",
34528
+ "input_cost_per_token": 0.0044,
34529
+ "output_cost_per_token": 0.0176,
34530
+ "max_input_tokens": 1000000,
34531
+ "max_output_tokens": 40000,
34532
+ "max_tokens": 40000,
34533
+ "supports_function_calling": true,
34534
+ "supports_parallel_function_calling": true,
34535
+ "supports_tool_choice": true,
34536
+ "supports_system_messages": true,
34537
+ "supports_reasoning": true
34538
+ },
34539
+ "novita/deepseek/deepseek-r1-0528": {
34540
+ "litellm_provider": "novita",
34541
+ "mode": "chat",
34542
+ "input_cost_per_token": 0.0056,
34543
+ "output_cost_per_token": 0.02,
34544
+ "max_input_tokens": 163840,
34545
+ "max_output_tokens": 32768,
34546
+ "max_tokens": 32768,
34547
+ "supports_function_calling": true,
34548
+ "supports_parallel_function_calling": true,
34549
+ "supports_tool_choice": true,
34550
+ "supports_system_messages": true,
34551
+ "supports_response_schema": true,
34552
+ "cache_read_input_token_cost": 0.0028,
34553
+ "input_cost_per_token_cache_hit": 0.0028,
34554
+ "supports_reasoning": true
34555
+ },
34556
+ "novita/deepseek/deepseek-r1-distill-qwen-32b": {
34557
+ "litellm_provider": "novita",
34558
+ "mode": "chat",
34559
+ "input_cost_per_token": 0.0024,
34560
+ "output_cost_per_token": 0.0024,
34561
+ "max_input_tokens": 64000,
34562
+ "max_output_tokens": 32000,
34563
+ "max_tokens": 32000,
34564
+ "supports_tool_choice": true,
34565
+ "supports_system_messages": true,
34566
+ "supports_response_schema": true,
34567
+ "supports_reasoning": true
34568
+ },
34569
+ "novita/meta-llama/llama-3-8b-instruct": {
34570
+ "litellm_provider": "novita",
34571
+ "mode": "chat",
34572
+ "input_cost_per_token": 0.00032,
34573
+ "output_cost_per_token": 0.00032,
34574
+ "max_input_tokens": 8192,
34575
+ "max_output_tokens": 8192,
34576
+ "max_tokens": 8192,
34577
+ "supports_system_messages": true
34578
+ },
34579
+ "novita/microsoft/wizardlm-2-8x22b": {
34580
+ "litellm_provider": "novita",
34581
+ "mode": "chat",
34582
+ "input_cost_per_token": 0.00496,
34583
+ "output_cost_per_token": 0.00496,
34584
+ "max_input_tokens": 65535,
34585
+ "max_output_tokens": 8000,
34586
+ "max_tokens": 8000,
34587
+ "supports_system_messages": true
34588
+ },
34589
+ "novita/deepseek/deepseek-r1-0528-qwen3-8b": {
34590
+ "litellm_provider": "novita",
34591
+ "mode": "chat",
34592
+ "input_cost_per_token": 0.00048,
34593
+ "output_cost_per_token": 0.00072,
34594
+ "max_input_tokens": 128000,
34595
+ "max_output_tokens": 32000,
34596
+ "max_tokens": 32000,
34597
+ "supports_system_messages": true,
34598
+ "supports_reasoning": true
34599
+ },
34600
+ "novita/deepseek/deepseek-r1-distill-llama-70b": {
34601
+ "litellm_provider": "novita",
34602
+ "mode": "chat",
34603
+ "input_cost_per_token": 0.0064,
34604
+ "output_cost_per_token": 0.0064,
34605
+ "max_input_tokens": 8192,
34606
+ "max_output_tokens": 8192,
34607
+ "max_tokens": 8192,
34608
+ "supports_tool_choice": true,
34609
+ "supports_system_messages": true,
34610
+ "supports_response_schema": true,
34611
+ "supports_reasoning": true
34612
+ },
34613
+ "novita/meta-llama/llama-3-70b-instruct": {
34614
+ "litellm_provider": "novita",
34615
+ "mode": "chat",
34616
+ "input_cost_per_token": 0.0051,
34617
+ "output_cost_per_token": 0.0074,
34618
+ "max_input_tokens": 8192,
34619
+ "max_output_tokens": 8000,
34620
+ "max_tokens": 8000,
34621
+ "supports_tool_choice": true,
34622
+ "supports_system_messages": true,
34623
+ "supports_response_schema": true
34624
+ },
34625
+ "novita/qwen/qwen3-235b-a22b-fp8": {
34626
+ "litellm_provider": "novita",
34627
+ "mode": "chat",
34628
+ "input_cost_per_token": 0.0016,
34629
+ "output_cost_per_token": 0.0064,
34630
+ "max_input_tokens": 40960,
34631
+ "max_output_tokens": 20000,
34632
+ "max_tokens": 20000,
34633
+ "supports_system_messages": true,
34634
+ "supports_reasoning": true
34635
+ },
34636
+ "novita/meta-llama/llama-4-maverick-17b-128e-instruct-fp8": {
34637
+ "litellm_provider": "novita",
34638
+ "mode": "chat",
34639
+ "input_cost_per_token": 0.0016,
34640
+ "output_cost_per_token": 0.0072,
34641
+ "max_input_tokens": 1048576,
34642
+ "max_output_tokens": 8192,
34643
+ "max_tokens": 8192,
34644
+ "supports_vision": true,
34645
+ "supports_system_messages": true
34646
+ },
34647
+ "novita/meta-llama/llama-4-scout-17b-16e-instruct": {
34648
+ "litellm_provider": "novita",
34649
+ "mode": "chat",
34650
+ "input_cost_per_token": 0.0008,
34651
+ "output_cost_per_token": 0.004,
34652
+ "max_input_tokens": 131072,
34653
+ "max_output_tokens": 131072,
34654
+ "max_tokens": 131072,
34655
+ "supports_vision": true,
34656
+ "supports_system_messages": true
34657
+ },
34658
+ "novita/nousresearch/hermes-2-pro-llama-3-8b": {
34659
+ "litellm_provider": "novita",
34660
+ "mode": "chat",
34661
+ "input_cost_per_token": 0.0014,
34662
+ "output_cost_per_token": 0.0014,
34663
+ "max_input_tokens": 8192,
34664
+ "max_output_tokens": 8192,
34665
+ "max_tokens": 8192,
34666
+ "supports_tool_choice": true,
34667
+ "supports_system_messages": true,
34668
+ "supports_response_schema": true
34669
+ },
34670
+ "novita/qwen/qwen2.5-vl-72b-instruct": {
34671
+ "litellm_provider": "novita",
34672
+ "mode": "chat",
34673
+ "input_cost_per_token": 0.0064,
34674
+ "output_cost_per_token": 0.0064,
34675
+ "max_input_tokens": 32768,
34676
+ "max_output_tokens": 32768,
34677
+ "max_tokens": 32768,
34678
+ "supports_vision": true,
34679
+ "supports_system_messages": true
34680
+ },
34681
+ "novita/sao10k/l3-70b-euryale-v2.1": {
34682
+ "litellm_provider": "novita",
34683
+ "mode": "chat",
34684
+ "input_cost_per_token": 0.0148,
34685
+ "output_cost_per_token": 0.0148,
34686
+ "max_input_tokens": 8192,
34687
+ "max_output_tokens": 8192,
34688
+ "max_tokens": 8192,
34689
+ "supports_function_calling": true,
34690
+ "supports_parallel_function_calling": true,
34691
+ "supports_tool_choice": true,
34692
+ "supports_system_messages": true
34693
+ },
34694
+ "novita/baidu/ernie-4.5-21B-a3b-thinking": {
34695
+ "litellm_provider": "novita",
34696
+ "mode": "chat",
34697
+ "input_cost_per_token": 0.00056,
34698
+ "output_cost_per_token": 0.00224,
34699
+ "max_input_tokens": 131072,
34700
+ "max_output_tokens": 65536,
34701
+ "max_tokens": 65536,
34702
+ "supports_system_messages": true,
34703
+ "supports_reasoning": true
34704
+ },
34705
+ "novita/sao10k/l3-8b-lunaris": {
34706
+ "litellm_provider": "novita",
34707
+ "mode": "chat",
34708
+ "input_cost_per_token": 0.0005,
34709
+ "output_cost_per_token": 0.0005,
34710
+ "max_input_tokens": 8192,
34711
+ "max_output_tokens": 8192,
34712
+ "max_tokens": 8192,
34713
+ "supports_tool_choice": true,
34714
+ "supports_system_messages": true,
34715
+ "supports_response_schema": true
34716
+ },
34717
+ "novita/baichuan/baichuan-m2-32b": {
34718
+ "litellm_provider": "novita",
34719
+ "mode": "chat",
34720
+ "input_cost_per_token": 0.00056,
34721
+ "output_cost_per_token": 0.00056,
34722
+ "max_input_tokens": 131072,
34723
+ "max_output_tokens": 131072,
34724
+ "max_tokens": 131072,
34725
+ "supports_tool_choice": true,
34726
+ "supports_system_messages": true,
34727
+ "supports_response_schema": true,
34728
+ "supports_reasoning": true
34729
+ },
34730
+ "novita/thudm/glm-4.1v-9b-thinking": {
34731
+ "litellm_provider": "novita",
34732
+ "mode": "chat",
34733
+ "input_cost_per_token": 0.00028,
34734
+ "output_cost_per_token": 0.001104,
34735
+ "max_input_tokens": 65536,
34736
+ "max_output_tokens": 8000,
34737
+ "max_tokens": 8000,
34738
+ "supports_vision": true,
34739
+ "supports_system_messages": true,
34740
+ "supports_reasoning": true
34741
+ },
34742
+ "novita/baidu/ernie-4.5-vl-424b-a47b": {
34743
+ "litellm_provider": "novita",
34744
+ "mode": "chat",
34745
+ "input_cost_per_token": 0.00336,
34746
+ "output_cost_per_token": 0.01,
34747
+ "max_input_tokens": 123000,
34748
+ "max_output_tokens": 16000,
34749
+ "max_tokens": 16000,
34750
+ "supports_vision": true,
34751
+ "supports_system_messages": true,
34752
+ "supports_reasoning": true
34753
+ },
34754
+ "novita/baidu/ernie-4.5-300b-a47b-paddle": {
34755
+ "litellm_provider": "novita",
34756
+ "mode": "chat",
34757
+ "input_cost_per_token": 0.00224,
34758
+ "output_cost_per_token": 0.0088,
34759
+ "max_input_tokens": 123000,
34760
+ "max_output_tokens": 12000,
34761
+ "max_tokens": 12000,
34762
+ "supports_tool_choice": true,
34763
+ "supports_system_messages": true,
34764
+ "supports_response_schema": true
34765
+ },
34766
+ "novita/deepseek/deepseek-prover-v2-671b": {
34767
+ "litellm_provider": "novita",
34768
+ "mode": "chat",
34769
+ "input_cost_per_token": 0.0056,
34770
+ "output_cost_per_token": 0.02,
34771
+ "max_input_tokens": 160000,
34772
+ "max_output_tokens": 160000,
34773
+ "max_tokens": 160000,
34774
+ "supports_system_messages": true
34775
+ },
34776
+ "novita/qwen/qwen3-32b-fp8": {
34777
+ "litellm_provider": "novita",
34778
+ "mode": "chat",
34779
+ "input_cost_per_token": 0.0008,
34780
+ "output_cost_per_token": 0.0036,
34781
+ "max_input_tokens": 40960,
34782
+ "max_output_tokens": 20000,
34783
+ "max_tokens": 20000,
34784
+ "supports_system_messages": true,
34785
+ "supports_reasoning": true
34786
+ },
34787
+ "novita/qwen/qwen3-30b-a3b-fp8": {
34788
+ "litellm_provider": "novita",
34789
+ "mode": "chat",
34790
+ "input_cost_per_token": 0.00072,
34791
+ "output_cost_per_token": 0.0036,
34792
+ "max_input_tokens": 40960,
34793
+ "max_output_tokens": 20000,
34794
+ "max_tokens": 20000,
34795
+ "supports_system_messages": true,
34796
+ "supports_reasoning": true
34797
+ },
34798
+ "novita/google/gemma-3-27b-it": {
34799
+ "litellm_provider": "novita",
34800
+ "mode": "chat",
34801
+ "input_cost_per_token": 0.000952,
34802
+ "output_cost_per_token": 0.0016,
34803
+ "max_input_tokens": 98304,
34804
+ "max_output_tokens": 16384,
34805
+ "max_tokens": 16384,
34806
+ "supports_vision": true,
34807
+ "supports_system_messages": true
34808
+ },
34809
+ "novita/deepseek/deepseek-v3-turbo": {
34810
+ "litellm_provider": "novita",
34811
+ "mode": "chat",
34812
+ "input_cost_per_token": 0.0032,
34813
+ "output_cost_per_token": 0.0104,
34814
+ "max_input_tokens": 64000,
34815
+ "max_output_tokens": 16000,
34816
+ "max_tokens": 16000,
34817
+ "supports_function_calling": true,
34818
+ "supports_parallel_function_calling": true,
34819
+ "supports_tool_choice": true,
34820
+ "supports_system_messages": true
34821
+ },
34822
+ "novita/deepseek/deepseek-r1-turbo": {
34823
+ "litellm_provider": "novita",
34824
+ "mode": "chat",
34825
+ "input_cost_per_token": 0.0056,
34826
+ "output_cost_per_token": 0.02,
34827
+ "max_input_tokens": 64000,
34828
+ "max_output_tokens": 16000,
34829
+ "max_tokens": 16000,
34830
+ "supports_function_calling": true,
34831
+ "supports_parallel_function_calling": true,
34832
+ "supports_tool_choice": true,
34833
+ "supports_system_messages": true,
34834
+ "supports_reasoning": true
34835
+ },
34836
+ "novita/Sao10K/L3-8B-Stheno-v3.2": {
34837
+ "litellm_provider": "novita",
34838
+ "mode": "chat",
34839
+ "input_cost_per_token": 0.0005,
34840
+ "output_cost_per_token": 0.0005,
34841
+ "max_input_tokens": 8192,
34842
+ "max_output_tokens": 32000,
34843
+ "max_tokens": 32000,
34844
+ "supports_function_calling": true,
34845
+ "supports_parallel_function_calling": true,
34846
+ "supports_tool_choice": true,
34847
+ "supports_system_messages": true
34848
+ },
34849
+ "novita/gryphe/mythomax-l2-13b": {
34850
+ "litellm_provider": "novita",
34851
+ "mode": "chat",
34852
+ "input_cost_per_token": 0.00072,
34853
+ "output_cost_per_token": 0.00072,
34854
+ "max_input_tokens": 4096,
34855
+ "max_output_tokens": 3200,
34856
+ "max_tokens": 3200,
34857
+ "supports_system_messages": true
34858
+ },
34859
+ "novita/baidu/ernie-4.5-vl-28b-a3b-thinking": {
34860
+ "litellm_provider": "novita",
34861
+ "mode": "chat",
34862
+ "input_cost_per_token": 0.0039,
34863
+ "output_cost_per_token": 0.0039,
34864
+ "max_input_tokens": 131072,
34865
+ "max_output_tokens": 65536,
34866
+ "max_tokens": 65536,
34867
+ "supports_function_calling": true,
34868
+ "supports_parallel_function_calling": true,
34869
+ "supports_tool_choice": true,
34870
+ "supports_vision": true,
34871
+ "supports_system_messages": true,
34872
+ "supports_response_schema": true,
34873
+ "supports_reasoning": true
34874
+ },
34875
+ "novita/qwen/qwen3-vl-8b-instruct": {
34876
+ "litellm_provider": "novita",
34877
+ "mode": "chat",
34878
+ "input_cost_per_token": 0.00064,
34879
+ "output_cost_per_token": 0.004,
34880
+ "max_input_tokens": 131072,
34881
+ "max_output_tokens": 32768,
34882
+ "max_tokens": 32768,
34883
+ "supports_function_calling": true,
34884
+ "supports_parallel_function_calling": true,
34885
+ "supports_tool_choice": true,
34886
+ "supports_vision": true,
34887
+ "supports_system_messages": true,
34888
+ "supports_response_schema": true
34889
+ },
34890
+ "novita/zai-org/glm-4.5-air": {
34891
+ "litellm_provider": "novita",
34892
+ "mode": "chat",
34893
+ "input_cost_per_token": 0.00104,
34894
+ "output_cost_per_token": 0.0068,
34895
+ "max_input_tokens": 131072,
34896
+ "max_output_tokens": 98304,
34897
+ "max_tokens": 98304,
34898
+ "supports_function_calling": true,
34899
+ "supports_parallel_function_calling": true,
34900
+ "supports_tool_choice": true,
34901
+ "supports_system_messages": true,
34902
+ "supports_reasoning": true
34903
+ },
34904
+ "novita/qwen/qwen3-vl-30b-a3b-instruct": {
34905
+ "litellm_provider": "novita",
34906
+ "mode": "chat",
34907
+ "input_cost_per_token": 0.0016,
34908
+ "output_cost_per_token": 0.0056,
34909
+ "max_input_tokens": 131072,
34910
+ "max_output_tokens": 32768,
34911
+ "max_tokens": 32768,
34912
+ "supports_function_calling": true,
34913
+ "supports_parallel_function_calling": true,
34914
+ "supports_tool_choice": true,
34915
+ "supports_vision": true,
34916
+ "supports_system_messages": true,
34917
+ "supports_response_schema": true
34918
+ },
34919
+ "novita/qwen/qwen3-vl-30b-a3b-thinking": {
34920
+ "litellm_provider": "novita",
34921
+ "mode": "chat",
34922
+ "input_cost_per_token": 0.0016,
34923
+ "output_cost_per_token": 0.008,
34924
+ "max_input_tokens": 131072,
34925
+ "max_output_tokens": 32768,
34926
+ "max_tokens": 32768,
34927
+ "supports_function_calling": true,
34928
+ "supports_parallel_function_calling": true,
34929
+ "supports_tool_choice": true,
34930
+ "supports_vision": true,
34931
+ "supports_system_messages": true,
34932
+ "supports_response_schema": true
34933
+ },
34934
+ "novita/qwen/qwen-mt-plus": {
34935
+ "litellm_provider": "novita",
34936
+ "mode": "chat",
34937
+ "input_cost_per_token": 0.002,
34938
+ "output_cost_per_token": 0.006,
34939
+ "max_input_tokens": 16384,
34940
+ "max_output_tokens": 8192,
34941
+ "max_tokens": 8192,
34942
+ "supports_system_messages": true
34943
+ },
34944
+ "novita/baidu/ernie-4.5-vl-28b-a3b": {
34945
+ "litellm_provider": "novita",
34946
+ "mode": "chat",
34947
+ "input_cost_per_token": 0.00112,
34948
+ "output_cost_per_token": 0.00448,
34949
+ "max_input_tokens": 30000,
34950
+ "max_output_tokens": 8000,
34951
+ "max_tokens": 8000,
34952
+ "supports_function_calling": true,
34953
+ "supports_parallel_function_calling": true,
34954
+ "supports_tool_choice": true,
34955
+ "supports_vision": true,
34956
+ "supports_system_messages": true,
34957
+ "supports_reasoning": true
34958
+ },
34959
+ "novita/baidu/ernie-4.5-21B-a3b": {
34960
+ "litellm_provider": "novita",
34961
+ "mode": "chat",
34962
+ "input_cost_per_token": 0.00056,
34963
+ "output_cost_per_token": 0.00224,
34964
+ "max_input_tokens": 120000,
34965
+ "max_output_tokens": 8000,
34966
+ "max_tokens": 8000,
34967
+ "supports_function_calling": true,
34968
+ "supports_parallel_function_calling": true,
34969
+ "supports_tool_choice": true,
34970
+ "supports_system_messages": true
34971
+ },
34972
+ "novita/qwen/qwen3-8b-fp8": {
34973
+ "litellm_provider": "novita",
34974
+ "mode": "chat",
34975
+ "input_cost_per_token": 0.00028,
34976
+ "output_cost_per_token": 0.001104,
34977
+ "max_input_tokens": 128000,
34978
+ "max_output_tokens": 20000,
34979
+ "max_tokens": 20000,
34980
+ "supports_system_messages": true,
34981
+ "supports_reasoning": true
34982
+ },
34983
+ "novita/qwen/qwen3-4b-fp8": {
34984
+ "litellm_provider": "novita",
34985
+ "mode": "chat",
34986
+ "input_cost_per_token": 0.00024,
34987
+ "output_cost_per_token": 0.00024,
34988
+ "max_input_tokens": 128000,
34989
+ "max_output_tokens": 20000,
34990
+ "max_tokens": 20000,
34991
+ "supports_system_messages": true,
34992
+ "supports_reasoning": true
34993
+ },
34994
+ "novita/qwen/qwen2.5-7b-instruct": {
34995
+ "litellm_provider": "novita",
34996
+ "mode": "chat",
34997
+ "input_cost_per_token": 0.00056,
34998
+ "output_cost_per_token": 0.00056,
34999
+ "max_input_tokens": 32000,
35000
+ "max_output_tokens": 32000,
35001
+ "max_tokens": 32000,
35002
+ "supports_function_calling": true,
35003
+ "supports_parallel_function_calling": true,
35004
+ "supports_tool_choice": true,
35005
+ "supports_system_messages": true,
35006
+ "supports_response_schema": true
35007
+ },
35008
+ "novita/meta-llama/llama-3.2-3b-instruct": {
35009
+ "litellm_provider": "novita",
35010
+ "mode": "chat",
35011
+ "input_cost_per_token": 0.00024,
35012
+ "output_cost_per_token": 0.0004,
35013
+ "max_input_tokens": 32768,
35014
+ "max_output_tokens": 32000,
35015
+ "max_tokens": 32000,
35016
+ "supports_function_calling": true,
35017
+ "supports_parallel_function_calling": true,
35018
+ "supports_tool_choice": true,
35019
+ "supports_system_messages": true
35020
+ },
35021
+ "novita/sao10k/l31-70b-euryale-v2.2": {
35022
+ "litellm_provider": "novita",
35023
+ "mode": "chat",
35024
+ "input_cost_per_token": 0.0148,
35025
+ "output_cost_per_token": 0.0148,
35026
+ "max_input_tokens": 8192,
35027
+ "max_output_tokens": 8192,
35028
+ "max_tokens": 8192,
35029
+ "supports_function_calling": true,
35030
+ "supports_parallel_function_calling": true,
35031
+ "supports_tool_choice": true,
35032
+ "supports_system_messages": true
35033
+ },
35034
+ "novita/qwen/qwen3-embedding-0.6b": {
35035
+ "litellm_provider": "novita",
35036
+ "mode": "embedding",
35037
+ "input_cost_per_token": 0.00056,
35038
+ "output_cost_per_token": 0,
35039
+ "max_input_tokens": 32768,
35040
+ "max_output_tokens": 32768,
35041
+ "max_tokens": 32768
35042
+ },
35043
+ "novita/qwen/qwen3-embedding-8b": {
35044
+ "litellm_provider": "novita",
35045
+ "mode": "embedding",
35046
+ "input_cost_per_token": 0.00056,
35047
+ "output_cost_per_token": 0,
35048
+ "max_input_tokens": 32768,
35049
+ "max_output_tokens": 4096,
35050
+ "max_tokens": 4096
35051
+ },
35052
+ "novita/baai/bge-m3": {
35053
+ "litellm_provider": "novita",
35054
+ "mode": "embedding",
35055
+ "input_cost_per_token": 0.0001,
35056
+ "output_cost_per_token": 0.0001,
35057
+ "max_input_tokens": 8192,
35058
+ "max_output_tokens": 96000,
35059
+ "max_tokens": 96000
35060
+ },
35061
+ "novita/qwen/qwen3-reranker-8b": {
35062
+ "litellm_provider": "novita",
35063
+ "mode": "rerank",
35064
+ "input_cost_per_token": 0.0004,
35065
+ "output_cost_per_token": 0.0004,
35066
+ "max_input_tokens": 32768,
35067
+ "max_output_tokens": 4096,
35068
+ "max_tokens": 4096
35069
+ },
35070
+ "novita/baai/bge-reranker-v2-m3": {
35071
+ "litellm_provider": "novita",
35072
+ "mode": "rerank",
35073
+ "input_cost_per_token": 0.0001,
35074
+ "output_cost_per_token": 0.0001,
35075
+ "max_input_tokens": 8000,
35076
+ "max_output_tokens": 8000,
35077
+ "max_tokens": 8000
35078
+ },
35079
+ "replicate/openai/gpt-5": {
35080
+ "input_cost_per_token": 1.25e-06,
35081
+ "output_cost_per_token": 1e-05,
35082
+ "litellm_provider": "replicate",
35083
+ "mode": "chat",
35084
+ "supports_function_calling": true,
35085
+ "supports_parallel_function_calling": true,
35086
+ "supports_vision": true,
35087
+ "supports_system_messages": true,
35088
+ "supports_tool_choice": true,
35089
+ "supports_response_schema": true
35090
+ },
35091
+ "replicateopenai/gpt-oss-20b": {
35092
+ "input_cost_per_token": 9e-08,
35093
+ "output_cost_per_token": 3.6e-07,
35094
+ "litellm_provider": "replicate",
35095
+ "mode": "chat",
35096
+ "supports_function_calling": true,
35097
+ "supports_system_messages": true
35098
+ },
35099
+ "replicate/anthropic/claude-4.5-haiku": {
35100
+ "input_cost_per_token": 1e-06,
35101
+ "output_cost_per_token": 5e-06,
35102
+ "litellm_provider": "replicate",
35103
+ "mode": "chat",
35104
+ "supports_function_calling": true,
35105
+ "supports_parallel_function_calling": true,
35106
+ "supports_vision": true,
35107
+ "supports_system_messages": true,
35108
+ "supports_tool_choice": true,
35109
+ "supports_response_schema": true,
35110
+ "supports_prompt_caching": true
35111
+ },
35112
+ "replicate/ibm-granite/granite-3.3-8b-instruct": {
35113
+ "input_cost_per_token": 3e-08,
35114
+ "output_cost_per_token": 2.5e-07,
35115
+ "litellm_provider": "replicate",
35116
+ "mode": "chat",
35117
+ "supports_function_calling": true,
35118
+ "supports_system_messages": true
35119
+ },
35120
+ "replicate/openai/gpt-4o": {
35121
+ "input_cost_per_token": 2.5e-06,
35122
+ "output_cost_per_token": 1e-05,
35123
+ "litellm_provider": "replicate",
35124
+ "mode": "chat",
35125
+ "supports_function_calling": true,
35126
+ "supports_parallel_function_calling": true,
35127
+ "supports_vision": true,
35128
+ "supports_system_messages": true,
35129
+ "supports_tool_choice": true,
35130
+ "supports_response_schema": true,
35131
+ "supports_audio_input": true,
35132
+ "supports_audio_output": true
35133
+ },
35134
+ "replicate/openai/o4-mini": {
35135
+ "input_cost_per_token": 1e-06,
35136
+ "output_cost_per_token": 4e-06,
35137
+ "output_cost_per_reasoning_token": 4e-06,
35138
+ "litellm_provider": "replicate",
35139
+ "mode": "chat",
35140
+ "supports_reasoning": true,
35141
+ "supports_system_messages": true
35142
+ },
35143
+ "replicate/openai/o1-mini": {
35144
+ "input_cost_per_token": 1.1e-06,
35145
+ "output_cost_per_token": 4.4e-06,
35146
+ "output_cost_per_reasoning_token": 4.4e-06,
35147
+ "litellm_provider": "replicate",
35148
+ "mode": "chat",
35149
+ "supports_reasoning": true,
35150
+ "supports_system_messages": true
35151
+ },
35152
+ "replicate/openai/o1": {
35153
+ "input_cost_per_token": 1.5e-05,
35154
+ "output_cost_per_token": 6e-05,
35155
+ "output_cost_per_reasoning_token": 6e-05,
35156
+ "litellm_provider": "replicate",
35157
+ "mode": "chat",
35158
+ "supports_reasoning": true,
35159
+ "supports_system_messages": true
35160
+ },
35161
+ "replicate/openai/gpt-4o-mini": {
35162
+ "input_cost_per_token": 1.5e-07,
35163
+ "output_cost_per_token": 6e-07,
35164
+ "litellm_provider": "replicate",
35165
+ "mode": "chat",
35166
+ "supports_function_calling": true,
35167
+ "supports_parallel_function_calling": true,
35168
+ "supports_vision": true,
35169
+ "supports_system_messages": true,
35170
+ "supports_tool_choice": true,
35171
+ "supports_response_schema": true
35172
+ },
35173
+ "replicate/qwen/qwen3-235b-a22b-instruct-2507": {
35174
+ "input_cost_per_token": 2.64e-07,
35175
+ "output_cost_per_token": 1.06e-06,
35176
+ "litellm_provider": "replicate",
35177
+ "mode": "chat",
35178
+ "supports_function_calling": true,
35179
+ "supports_system_messages": true
35180
+ },
35181
+ "replicate/anthropic/claude-4-sonnet": {
35182
+ "input_cost_per_token": 3e-06,
35183
+ "output_cost_per_token": 1.5e-05,
35184
+ "litellm_provider": "replicate",
35185
+ "mode": "chat",
35186
+ "supports_function_calling": true,
35187
+ "supports_parallel_function_calling": true,
35188
+ "supports_vision": true,
35189
+ "supports_system_messages": true,
35190
+ "supports_tool_choice": true,
35191
+ "supports_response_schema": true,
35192
+ "supports_prompt_caching": true
35193
+ },
35194
+ "replicate/deepseek-ai/deepseek-v3": {
35195
+ "input_cost_per_token": 1.45e-06,
35196
+ "output_cost_per_token": 1.45e-06,
35197
+ "litellm_provider": "replicate",
35198
+ "mode": "chat",
35199
+ "max_input_tokens": 65536,
35200
+ "max_output_tokens": 8192,
35201
+ "max_tokens": 8192,
35202
+ "supports_function_calling": true,
35203
+ "supports_system_messages": true
35204
+ },
35205
+ "replicate/anthropic/claude-3.7-sonnet": {
35206
+ "input_cost_per_token": 3e-06,
35207
+ "output_cost_per_token": 1.5e-05,
35208
+ "litellm_provider": "replicate",
35209
+ "mode": "chat",
35210
+ "supports_function_calling": true,
35211
+ "supports_parallel_function_calling": true,
35212
+ "supports_vision": true,
35213
+ "supports_system_messages": true,
35214
+ "supports_tool_choice": true,
35215
+ "supports_response_schema": true,
35216
+ "supports_prompt_caching": true
35217
+ },
35218
+ "replicate/anthropic/claude-3.5-haiku": {
35219
+ "input_cost_per_token": 1e-06,
35220
+ "output_cost_per_token": 5e-06,
35221
+ "litellm_provider": "replicate",
35222
+ "mode": "chat",
35223
+ "supports_function_calling": true,
35224
+ "supports_parallel_function_calling": true,
35225
+ "supports_vision": true,
35226
+ "supports_system_messages": true,
35227
+ "supports_tool_choice": true,
35228
+ "supports_response_schema": true,
35229
+ "supports_prompt_caching": true
35230
+ },
35231
+ "replicate/anthropic/claude-3.5-sonnet": {
35232
+ "input_cost_per_token": 3.75e-06,
35233
+ "output_cost_per_token": 1.875e-05,
35234
+ "litellm_provider": "replicate",
35235
+ "mode": "chat",
35236
+ "supports_function_calling": true,
35237
+ "supports_parallel_function_calling": true,
35238
+ "supports_vision": true,
35239
+ "supports_system_messages": true,
35240
+ "supports_tool_choice": true,
35241
+ "supports_response_schema": true,
35242
+ "supports_prompt_caching": true
35243
+ },
35244
+ "replicate/google/gemini-3-pro": {
35245
+ "input_cost_per_token": 2e-06,
35246
+ "output_cost_per_token": 1.2e-05,
35247
+ "litellm_provider": "replicate",
35248
+ "mode": "chat",
35249
+ "supports_function_calling": true,
35250
+ "supports_parallel_function_calling": true,
35251
+ "supports_vision": true,
35252
+ "supports_system_messages": true,
35253
+ "supports_tool_choice": true,
35254
+ "supports_response_schema": true
35255
+ },
35256
+ "replicate/anthropic/claude-4.5-sonnet": {
35257
+ "input_cost_per_token": 3e-06,
35258
+ "output_cost_per_token": 1.5e-05,
35259
+ "litellm_provider": "replicate",
35260
+ "mode": "chat",
35261
+ "supports_function_calling": true,
35262
+ "supports_parallel_function_calling": true,
35263
+ "supports_vision": true,
35264
+ "supports_system_messages": true,
35265
+ "supports_tool_choice": true,
35266
+ "supports_response_schema": true,
35267
+ "supports_prompt_caching": true
35268
+ },
35269
+ "replicate/openai/gpt-4.1": {
35270
+ "input_cost_per_token": 2e-06,
35271
+ "output_cost_per_token": 8e-06,
35272
+ "litellm_provider": "replicate",
35273
+ "mode": "chat",
35274
+ "supports_function_calling": true,
35275
+ "supports_parallel_function_calling": true,
35276
+ "supports_vision": true,
35277
+ "supports_system_messages": true,
35278
+ "supports_tool_choice": true,
35279
+ "supports_response_schema": true
35280
+ },
35281
+ "replicate/openai/gpt-4.1-nano": {
35282
+ "input_cost_per_token": 1e-07,
35283
+ "output_cost_per_token": 4e-07,
35284
+ "litellm_provider": "replicate",
35285
+ "mode": "chat",
35286
+ "supports_function_calling": true,
35287
+ "supports_system_messages": true
35288
+ },
35289
+ "replicate/openai/gpt-4.1-mini": {
35290
+ "input_cost_per_token": 4e-07,
35291
+ "output_cost_per_token": 1.6e-06,
35292
+ "litellm_provider": "replicate",
35293
+ "mode": "chat",
35294
+ "supports_function_calling": true,
35295
+ "supports_parallel_function_calling": true,
35296
+ "supports_vision": true,
35297
+ "supports_system_messages": true,
35298
+ "supports_tool_choice": true,
35299
+ "supports_response_schema": true
35300
+ },
35301
+ "replicate/openai/gpt-5-nano": {
35302
+ "input_cost_per_token": 5e-08,
35303
+ "output_cost_per_token": 4e-07,
35304
+ "litellm_provider": "replicate",
35305
+ "mode": "chat",
35306
+ "supports_function_calling": true,
35307
+ "supports_system_messages": true
35308
+ },
35309
+ "replicate/openai/gpt-5-mini": {
35310
+ "input_cost_per_token": 2.5e-07,
35311
+ "output_cost_per_token": 2e-06,
35312
+ "litellm_provider": "replicate",
35313
+ "mode": "chat",
35314
+ "supports_function_calling": true,
35315
+ "supports_parallel_function_calling": true,
35316
+ "supports_vision": true,
35317
+ "supports_system_messages": true,
35318
+ "supports_tool_choice": true,
35319
+ "supports_response_schema": true
35320
+ },
35321
+ "replicate/google/gemini-2.5-flash": {
35322
+ "input_cost_per_token": 2.5e-06,
35323
+ "output_cost_per_token": 2.5e-06,
35324
+ "litellm_provider": "replicate",
35325
+ "mode": "chat",
35326
+ "supports_function_calling": true,
35327
+ "supports_parallel_function_calling": true,
35328
+ "supports_vision": true,
35329
+ "supports_system_messages": true,
35330
+ "supports_tool_choice": true,
35331
+ "supports_response_schema": true
35332
+ },
35333
+ "replicate/openai/gpt-oss-120b": {
35334
+ "input_cost_per_token": 1.8e-07,
35335
+ "output_cost_per_token": 7.2e-07,
35336
+ "litellm_provider": "replicate",
35337
+ "mode": "chat",
35338
+ "supports_function_calling": true,
35339
+ "supports_system_messages": true
35340
+ },
35341
+ "replicate/deepseek-ai/deepseek-v3.1": {
35342
+ "input_cost_per_token": 6.72e-07,
35343
+ "output_cost_per_token": 2.016e-06,
35344
+ "litellm_provider": "replicate",
35345
+ "mode": "chat",
35346
+ "max_input_tokens": 163840,
35347
+ "max_output_tokens": 163840,
35348
+ "max_tokens": 163840,
35349
+ "supports_function_calling": true,
35350
+ "supports_reasoning": true,
35351
+ "supports_system_messages": true
35352
+ },
35353
+ "replicate/xai/grok-4": {
35354
+ "input_cost_per_token": 7.2e-06,
35355
+ "output_cost_per_token": 3.6e-05,
35356
+ "litellm_provider": "replicate",
35357
+ "mode": "chat",
35358
+ "supports_function_calling": true,
35359
+ "supports_system_messages": true
35360
+ },
35361
+ "replicate/deepseek-ai/deepseek-r1": {
35362
+ "input_cost_per_token": 3.75e-06,
35363
+ "output_cost_per_token": 1e-05,
35364
+ "output_cost_per_reasoning_token": 1e-05,
35365
+ "litellm_provider": "replicate",
35366
+ "mode": "chat",
35367
+ "max_input_tokens": 65536,
35368
+ "max_output_tokens": 8192,
35369
+ "max_tokens": 8192,
35370
+ "supports_reasoning": true,
35371
+ "supports_system_messages": true
33243
35372
  }
33244
35373
  }